JPH09325788A - Device and method for voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To simply increase the variety of the voice qualities of synthesized voices without recording the uttered voice of an announcer and resegmenting speech elements. SOLUTION: The voice parameters, which are obtained by analyzing the discrete voice signals that are made into specimens by a first specimen period, are accumulated into a memory 11 and the fundamental frequency patterns of the voices are stored in a memory 12. Then, discrete voice signals are synthesized by a synthesis filter processing section 13 from the parameters read from the memory 11 and the patterns read from the memory 12. Then, a D/A converter 14 converts the discrete voice signals into analog voice signals by a second specimen period determined by a voice quality control section 18 in accordance with the voice quality that is switched and specified by a voice quality switching section 17. The analog voice signals are amplified by an amplifier 15 and outputted as voices from a speaker 16.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing apparatus suitable for parameterizing and storing a voice and synthesizing the voice from this, or subdividing and storing the voice and synthesizing an arbitrary voice by a combination thereof. And method.

[0002]

2. Description of the Related Art As a voice synthesizing device according to the present invention, a voice is parameterized and stored, and an analysis and synthesis device for synthesizing the voice from this, and a voice is subdivided and stored, and any voice can be synthesized by the combination. It is known that there is a rule synthesizer. In the following, examples of conventional techniques of the analysis and synthesis apparatus and the rule synthesis apparatus will be described with reference to the drawings.

FIG. 23 is a diagram showing the configuration of a conventional analysis and synthesis apparatus.

In FIG. 23, a memory 311 stores a time series of parameter frames obtained by analyzing the voice and voiced / unvoiced voice information corresponding to each frame. Here, a low-order cepstrum coefficient is used as a parameter.

By the way, the low-order cepstrum coefficient can be obtained as follows. First, a window function (here, a Hanning window) is applied to voice data uttered by an announcer or the like with a constant width and a constant period, and a Fourier transform is performed on the voice waveform in each window to calculate a short-time spectrum of the voice. Next, after the obtained short-time spectrum power is logarithmically obtained to obtain a logarithmic power spectrum, the logarithmic power spectrum is subjected to inverse Fourier transform. The cepstrum coefficient is calculated in this way.

It is generally known that a high-order cepstrum coefficient holds fundamental frequency information of a voice and a low-order cepstrum coefficient holds spectral envelope information of a voice.
Of these, only the low-order cepstrum representing the spectrum envelope of the voice is stored in the memory 311.

On the other hand, the memory 312 stores a time-series pattern of the fundamental frequency of the voice, which is obtained from the same voice by a predetermined fundamental frequency extraction method. As a fundamental frequency extraction method, for example, the document "" An by Bruce G et al.
Integrated Pitch TrackingAlgorithm for Speech Syst
Examples include methods such as ems ", ICASSP 1993".

The synthesis filter processing unit 313 reads out the above-mentioned respective data from these two memories 311 and 312, and generates a periodic pulse based on the time series pattern of the fundamental frequency in the voiced section and a random noise in the unvoiced section. As, the filter coefficient is calculated from the time series of the parameter frame (cepstral) of the voice and is given to the synthesis filter to synthesize the desired voice. Normally, an LMA (Log Magnitude Approximation) filter having a cepstrum coefficient as a direct filter coefficient is used as the synthesis filter. The processing up to this point is generally performed by a program.

Since the speech synthesized as described above is a discrete signal, the synthesis filter processing unit 313 finally converts this discrete signal (discrete waveform) into a D / A (digital / analog) converter 314. Supply to. In response to this, the D / A converter 314 converts the discrete signal (discrete audio signal) into an electrical analog signal (analog audio signal). By driving the speaker 316 and the like via the amplifier 315 with the analog signal obtained in this way, a sound that can be perceived aurally is synthesized.

FIG. 24 is a diagram showing the configuration of a conventional rule synthesizing device.

The rule-synthesizing device shown in FIG. 24 converts text into a symbol string consisting of phonemes and prosody, and generates speech from the symbol string (Text-to-speech conversion: T hereinafter).
(TS) processing is performed.

The TTS in the rule synthesizer of FIG.
The processing is roughly divided into a language processing unit 411 and a speech synthesis unit 4.
It is composed of two processing units 12 and is generally performed as follows, taking Japanese rule composition as an example.

First, the language processing unit 411 adds language processing such as morphological analysis and syntactic analysis to text (kanji / kana mixed sentences) input from the text file 413, decomposes into morphemes, estimates dependency relations, etc. At the same time, the reading and accent type are given to each morpheme.
After that, the language processing unit 412 determines an accent type for each phrase (hereinafter, referred to as an accent phrase) which is a delimiter at the time of reading by using an accent movement rule such as a compound word with respect to an accent. Normally, in the language processing section of TTS,
The reading and accent type for each accent phrase thus obtained can be output as a symbol string (hereinafter referred to as a phonetic symbol string).

Next, in the voice synthesizing unit 412, the duration of each phoneme included in the obtained reading is given to the phoneme duration calculating unit 414 according to a predetermined rule based on the phoneme environment of the phoneme. To decide. Then, according to the "reading" and the "phoneme duration" obtained as described above, the phoneme parameter generation processing unit 415 sequentially reads out the necessary speech units from the speech unit memory 416 and connects the read speech units. Then, the characteristic parameter sequence of the voice to be synthesized is generated.

Here, the speech unit memory 415 is composed of a large number of speech units created in advance. The speech unit analyzes the voice uttered by an announcer or the like to obtain a predetermined voice characteristic parameter, and then, in a predetermined synthesis unit, for example, a Japanese syllable (consonant + vowel: hereinafter referred to as CV) unit in Japanese. It is created by cutting out all the syllables included in the voice of from the characteristic parameter.

In the voice synthesizing unit 412, the pitch generation processing unit 417 further sets the point pitch at the time when the pitch changes based on the accent type, and linearly interpolates between the plurality of set point pitches. Generates a pitch accent component and superimposes an intonation component (normally a monotonically decreasing straight line on the frequency axis) on this to generate a pitch pattern. Then, the periodic pulse based on the pitch pattern is used as the sound source in the voiced section, and the random noise is used as the sound source in the unvoiced section, and the filter coefficient is calculated from the feature parameter sequence of one voice and is given to the synthesis filter processing unit 418 to synthesize the desired voice. .

Since the processing up to this point is generally performed by a program, and thus the synthesized speech is a discrete signal, the speech synthesis section 412 finally supplies this discrete signal to the D / A converter 419. To do. In response to this, D
The / A converter 419 converts the discrete signal (discrete audio signal) into an electrical analog signal (analog audio signal). By driving the speaker 421 and the like via the amplifier 420 with the analog signal thus obtained, aurally perceptible voice can be synthesized.

[0018]

[Problems to be Solved by the Invention]
At present, the above-mentioned conventional techniques exist, but the speech synthesized by this conventional speech synthesizer has the following problems. That is, in the conventional voice synthesizer, there is a restriction on the type of voice of the synthesized voice (hereinafter, referred to as voice quality). Can only be synthesized with. Therefore, when trying to increase the voice quality of synthesized speech when synthesizing a spoken sentence, the speech synthesizer developer newly hires a different announcer, records the utterance,
The creation of the speech unit has to be restarted from the beginning.
For this reason, wages are required to hire an announcer,
In addition, the developer requires a great deal of work for recording the voice of the announcer and cutting out the voice unit. And this leads to an increase in the cost of device development.

The present invention has been made in consideration of such circumstances, and an object thereof is to record a voice of an announcer and to re-cut a voice segment without using an extremely easy means. It is an object of the present invention to provide a voice synthesizing device and method capable of increasing the number of voices.

[0020]

A speech synthesizer according to a first aspect of the present invention is storage means for storing characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period. And a synthesizing means for synthesizing the discrete speech signal with the characteristic parameter of the speech read out from the accumulating means as an input, and a second sampling period different from the first sampling period for the discrete speech signal synthesized by the synthesizing means. And a digital / analog conversion means for converting into an analog voice signal.

In addition to the above, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided,
The analog converting means may be configured such that the discrete voice signal synthesized by the synthesizing means is converted into an analog voice signal at a second sample period determined according to the voice quality selected and designated by the voice quality selecting means. .

In the present invention, the first sampling period, which is the sampling period during voice analysis, and the digital / analog (D /
A) Since the spectrum of the voice is shifted on the frequency logarithmic axis by making the second sampling period in the conversion different, it is possible to synthesize voices of different voice qualities using the same feature parameter.

Next, a speech synthesizer according to a second aspect of the present invention is a speech obtained by analyzing a discrete speech signal sampled at a first sampling period by applying a time window at a first frame period and analyzing it. From the time series of the characteristic parameter frames of 1., a synthesizing means for synthesizing a discrete speech signal in a second frame cycle different from the first frame cycle, and a discrete speech signal synthesized by this synthesizing means in the first sample. And a digital / analog conversion means for converting into an analog audio signal at a second sampling period different from the period. Here, the second frame period is the first frame period,
A first sampling period and a second sampling period different from the first sampling period;
It should be determined based on the sampling period of.

In addition to the above, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and the synthesis means discretely in the second frame period determined according to the voice quality selected and designated by the voice quality selection means. The voice signals are synthesized, and in the digital / analog converting means, the discrete voice signals synthesized by the synthesizing means are transformed into analog voice signals at the second sample period determined according to the voice quality selected and designated by the voice quality selecting means. It is also possible to adopt a configuration.

According to the present invention, the spectrum of voice is shifted on the frequency logarithmic axis by making the first sampling period at the time of speech analysis different from the second sampling period at D / A conversion, and the same characteristic parameter is obtained. Can be used to synthesize voices of different voice qualities, and here, the second frame period during synthesis is determined based on the first frame period, the first sampling period, and the second sampling period. Alternatively, the second frame period during analysis and the second sample period during D / A conversion are determined according to the voice quality selected and designated by the voice quality selecting means. By making it different from the window cycle (first frame cycle), the speech rate of the synthesized speech can be controlled appropriately, so that speech with a natural speech rate can be synthesized while changing the voice quality.

Next, the speech synthesis apparatus according to the third aspect of the present invention comprises a characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at the first sampling period. A pitch pattern accumulating means for accumulating a fundamental frequency pattern of voice and a fundamental frequency pattern read from the pitch pattern accumulating means based on the first sampling period and a second sampling period different from the first sampling period. Pitch pattern modulating means for modulating, synthesizing means for synthesizing a discrete voice signal from the characteristic parameters read from the characteristic parameter accumulating means and the fundamental frequency pattern modulated by the pitch pattern modulating means, and the synthesizing means. The discrete voice signal
Digital / conversion to analog voice signal at sampling period of
And an analog conversion means.

In addition to the above, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and the pitch pattern modulation means modulates the fundamental frequency pattern according to the voice quality selected and designated by the voice quality selection means. Is done
In the digital / analog converting means, the discrete voice signal synthesized by the synthesizing means may be converted into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means. It is possible.

In the present invention, the spectrum of voice is shifted on the frequency logarithmic axis by making the first sampling period, which is the sampling period at the time of speech analysis, different from the second sampling period of D / A conversion. Although voices having different voice qualities can be synthesized using the same characteristic parameter, the fundamental frequency pattern for synthesis can be determined based on the first sampling period and the second sampling period, or selected by the voice quality selecting means. Determine the second sampling period according to the specified voice quality,
By determining the fundamental frequency pattern for synthesis according to the selected and designated voice quality, the pitch of the voice to be synthesized (voice pitch) can be appropriately controlled, so that a voice with a natural pitch can be synthesized while changing the voice quality.

Next, the speech synthesizer according to the fourth aspect of the present invention selects a speech unit created from the discrete speech signal sampled at the first sampling period based on the given phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting it, and a digital / digital converter for converting the discrete voice signal synthesized by the voice synthesizing unit into an analog voice signal at a second sampling period different from the first sampling period. And an analog conversion means.

In addition, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and the digital / digital
The analog converting means may be configured such that the discrete voice signal synthesized by the synthesizing means is converted into an analog voice signal at a second sample period determined according to the voice quality selected and designated by the voice quality selecting means. .

In the present invention, since the first sampling period at the time of creating a speech unit and the second sampling period of D / A conversion are made different, the spectrum of the voice shifts on the frequency logarithmic axis. It is possible to synthesize voices having different voice qualities by using the voice unit.

Next, the speech synthesizer according to the fifth aspect of the present invention selects a speech unit created from the discrete speech signal sampled at the first sampling period based on the given phoneme information. A speech synthesizing means for synthesizing a discrete speech signal by connecting the selected speech unit according to a speech rate parameter relating to a speech rate or a speech time of a speech to be synthesized. Based on the first sampling period and a second sampling period different from the first sampling period, and a discrete speech signal synthesized by the speech synthesizing unit in the second sampling period. And a digital / analog converting means for converting into an analog audio signal.

In addition, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided. In the voice synthesis means, the speech speed parameter to be used is determined according to the voice quality selected and designated by the voice quality selection means. In the digital / analog converting means, the discrete voice signal synthesized by the voice synthesizing means is converted into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means. It is also possible.

In the present invention, the first sampling period at the time of creating a speech unit and the second sampling period of D / A conversion are made different to shift the spectrum of the speech on the frequency logarithmic axis, and the same characteristic is obtained. Although voices having different voice qualities can be synthesized using the parameters, here, the speech rate parameter used at the time of synthesis is determined according to the first sampling period and the second sampling period, or the voice quality selecting means is used. The second sampling period is determined according to the voice quality selected and designated by
By determining the speech rate parameter according to the selected and designated voice quality, the speech rate of the synthesized speech can be controlled appropriately, so that the speech can be synthesized at a natural speech rate while changing the voice quality.

Next, the speech synthesizer according to the sixth aspect of the present invention determines the duration of each phoneme included in the given phoneme information, while the discrete speech signal sampled at the first sampling period. A speech synthesizing unit that synthesizes a discrete speech signal by selecting a speech unit created from the above based on the phoneme information, and connecting the selected speech units based on the duration of each phoneme determined above. , A voice synthesizing unit that determines the phoneme duration to be used based on the first sampling period and a second sampling period different from the first sampling period, and an analog discrete voice signal synthesized by the speech synthesizing unit. And a digital / analog conversion means for converting into a voice signal.

In addition to this, voice quality selection means for selecting and designating the voice quality of the voice to be synthesized is further provided, and in the voice synthesis means, the phoneme duration to be used is determined according to the voice quality selected and designated by the voice quality selection means. In the digital / analog converting means, the discrete voice signal synthesized by the voice synthesizing means is converted into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means. It is also possible.

According to the present invention, the first sampling period at the time of creating a speech segment is made different from the second sampling period of D / A conversion to shift the speech spectrum on the frequency logarithmic axis, and the same characteristics are obtained. Voices of different voice qualities can be synthesized using the parameters, and here, the phoneme duration at the time of synthesis is determined according to the first sampling period and the second sampling period, or selected and designated by the voice quality selection means. By determining the second sampling period according to the voice quality that is generated and determining the phoneme duration according to the selected and specified voice quality, the utterance speed of the synthesized voice can be appropriately controlled, so that the voice quality can be changed naturally. Speech can be synthesized at the speaking rate.

Next, the speech synthesizer according to the seventh aspect of the present invention selects a speech unit created from the discrete speech signal sampled at the first sampling period based on the given phoneme information. The speech synthesis means for synthesizing the discrete speech signal by connecting the selected speech segments while expanding and contracting in the time axis direction, and the discrete sampling speech signal synthesized by the speech synthesizing means as the first sampling period. And a digital / analog conversion means for converting into an analog voice signal at a different second sampling period. Here, the degree of expansion / contraction in the time axis direction with respect to the voice unit when the voice unit is connected in the voice synthesizing unit may be determined based on the first sampling period and the second sampling period.

In addition to this, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided. In the voice synthesis means, the selected voice segment is determined according to the voice quality selected and designated by the voice quality selection means. In the digital / analog converting means, the discrete voice signals synthesized by the voice synthesizing means are determined in accordance with the voice quality selected and designated by the voice quality selecting means. It is also possible to adopt a configuration in which it is converted into an analog audio signal at a sampling period of 2.

In the present invention, the first sampling period at the time of creating a speech unit and the second sampling period of D / A conversion are made different to shift the spectrum of the speech on the frequency logarithmic axis, and the same characteristic is obtained. Although it is possible to synthesize voices having different voice qualities by using the parameters, the speech unit is expanded or contracted in the time axis direction at a degree determined based on the first sampling period and the second sampling period. The second sampling period is determined according to the voice quality selected and designated by the voice quality selection means, and the voice unit is connected while expanding and contracting in the time axis direction to the extent determined according to the voice quality selected and designated. By doing so, it is possible to properly control the speech rate of the synthesized speech and also to properly control the temporal change of the spectrum transient part of the synthesized speech, so that the natural speech can be made while changing the voice quality. Clear voice speed can be synthesized.

Next, the speech synthesis apparatus according to the eighth aspect of the present invention is a speech obtained by analyzing a discrete speech signal sampled at the first sampling period by applying a time window of the first frame period to the analysis. And a speech unit accumulating unit for accumulating a plurality of speech units cut out in a predetermined synthesis unit from the time series of the characteristic parameter frame, and selecting and connecting the speech units from the speech unit accumulating unit based on the input phoneme information. And a second parameter different from the first frame period based on the time series of the combined parameter frames generated by the combined parameter frame time series generation means and the combined parameter frame time series generation means for generating the time series of the combined parameter frames. A synthesizing means for synthesizing the discrete speech signal in the frame period of, and a discrete speech signal synthesized by the synthesizing means for the second sample different from the first sampling period. Characterized by comprising a digital / analog converting means for converting an analog audio signal in the cycle. Here, the second frame period may be determined based on the first frame period, the first sampling period, and the second sampling period different from the first sampling period.

In addition, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and the synthesis means discretely in the second frame cycle determined according to the voice quality selected and designated by the voice quality selection means. The voice signals are synthesized, and in the digital / analog converting means, the discrete voice signals synthesized by the synthesizing means are transformed into analog voice signals at the second sample period determined according to the voice quality selected and designated by the voice quality selecting means. It is also possible to adopt a configuration.

According to the present invention, the first sampling period at the time of creating a speech unit is made different from the second sampling period of D / A conversion to shift the spectrum of the speech on the frequency logarithmic axis, and the same characteristic is obtained. The parameters can be used to synthesize voices of different voice qualities, and further, here, the second frame period at the time of synthesis is based on the first frame period, the first sampling period, and the second sampling period. When the second frame period is analyzed, the second frame period during synthesis and the second sample period during D / A conversion are determined according to the voice quality selected and designated by the voice quality selection means. By making it different from the window period (first frame period), the speech rate of the speech to be synthesized can be controlled appropriately, so that speech with a natural speech rate can be synthesized while changing the voice quality.

Next, the speech synthesizer according to the ninth aspect of the present invention stores a plurality of speech units created from discrete speech signals sampled at the first sampling period with prosody information and phonological information as inputs. A speech unit accumulating means, a pitch pattern generating means for generating a fundamental frequency pattern of speech from the prosody information based on the first sampling period and a second sampling period different from the first sampling period, and the phoneme. Phonological parameter generation means for generating phonological parameter of the voice by selectively reading and connecting the speech element from the speech element accumulating means based on information, the phonological parameter generated by the phonological parameter generation means, and the pitch. A synthesizing means for synthesizing the discrete speech signal from the fundamental frequency pattern generated by the pattern generating means, and a separation means synthesized by this synthesizing means. Characterized in that the audio signal and a digital / analog converting means for converting an analog audio signal at the second sampling period.

In addition to this, voice quality selection means for selecting and designating the voice quality of the voice to be synthesized is further provided, and the pitch pattern generating means uses the prosody information to select the fundamental frequency of the voice in accordance with the voice quality selected and designated by the voice quality selection means. A pattern is generated, and in the digital / analog conversion means, the discrete voice signal synthesized by the synthesis means is converted into an analog voice signal at the second sampling period determined according to the voice quality selected and designated by the voice quality selection means. It can also be configured.

According to the present invention, the first sampling period at the time of creating a speech segment is made different from the second sampling period of D / A conversion, whereby the speech spectrum is shifted on the frequency logarithmic axis, and the same characteristic is obtained. Although voices having different voice qualities can be synthesized by using the parameters, here, the fundamental frequency pattern for synthesis is determined based on the first sampling period and the second sampling period, or the fundamental frequency pattern and D / A are used. By determining the second sampling period at the time of conversion according to the voice quality selected and designated by the voice quality selecting means, the pitch (voice pitch) of the synthesized voice can be appropriately controlled, and thus the voice quality can be changed naturally. Pitch voice can be synthesized.

Next, a voice synthesizing apparatus according to a tenth aspect of the present invention corresponds to the voice synthesizing apparatus according to the first aspect, and is a target of D / A conversion by digital / analog conversion means. And a sampling period converting means for converting the sampling period of the discrete speech signal into a second sampling period different from the first sampling period, and the digital-to-analog converting means converts the discrete speech signal into an analog speech signal. It is characterized in that the conversion is performed in a third cycle different from the second sampling cycle.

In addition, voice quality selection means for selecting and designating the voice quality of the voice to be synthesized is further provided, and in the sampling period conversion means, the sampling period of the discrete speech signal synthesized by the synthesis means is selected and designated by the voice quality selection means. It is also possible to adopt a configuration in which conversion is performed into the second sampling period that is determined according to the voice quality that is generated.

According to the present invention, after the synthesized speech (discrete speech signal) is converted into the second sampling period different from the first sampling period at the time of speech analysis, the third sampling period different from the second sampling period is obtained.
By performing D / A conversion at the sampling period of, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices of different voice qualities can be synthesized using the same feature parameter. In order to synthesize the voice with the original voice quality, the first sampling period = the second sampling period = the third sampling period.

Next, a speech synthesizer according to an eleventh aspect of the present invention corresponds to the speech synthesizer according to the second aspect, and is an object of D / A conversion by digital / analog conversion means. And a sampling period converting means for converting the sampling period of the discrete speech signal into a second sampling period different from the first sampling period, and the digital-to-analog converting means converts the discrete speech signal into an analog speech signal. It is characterized in that the conversion is performed in a third cycle different from the second sampling cycle. Here, the second frame period for synthesizing the discrete speech signal is
It may be determined based on the sampling period of 3 and the third sampling period.

In addition to the above, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and in the synthesis means, the discrete voice is generated in the second frame period determined according to the voice quality selected and designated by the voice quality selection means. The signals may be combined, and the sample conversion unit may be configured to convert the discrete voice signal combined by the combination unit into a second sample period determined according to the voice quality selected and designated by the voice quality selection unit. is there.

According to the present invention, after converting the synthesized speech into the second sampling period different from the first sampling period at the time of speech analysis, the D / D is converted into the third sampling period different from the second sampling period.
By performing the A conversion, the effect of shifting the spectrum of the voice on the frequency logarithmic axis can be obtained. Therefore, it is possible to synthesize voices having different voice qualities using the same feature parameter. The second frame cycle is determined based on the first frame cycle, the second sampling cycle, and the third sampling cycle, or the second sampling cycle and the second sampling cycle are combined.
Of the second frame period different from the window period (first frame period) at the time of analysis by, for example, determining the frame period of the above according to the voice quality selected and designated by the voice quality selecting means. Since it is possible to appropriately control the speech rate of, it is possible to synthesize a voice with a natural speech rate while changing the voice quality.

Next, the speech synthesizer according to the twelfth aspect of the present invention corresponds to the speech synthesizer according to the third aspect, in which the fundamental frequency pattern read from the pitch pattern accumulating means is the first frequency pattern. Pitch pattern modulating means for modulating based on a sampling period, a second sampling period different from the first sampling period and a third sampling period different from the second sampling period, and D / in the digital / analog converting means A sampling period conversion means for converting the sampling period of the discrete speech signal to be A-converted into the second sampling period, and conversion from the discrete speech signal to the analog speech signal in the digital / analog conversion means. Is performed in the above-mentioned third cycle.

In addition to this, voice quality selection means for selecting and specifying the voice quality of the voice to be synthesized is further provided, and the pitch pattern modulation means selects and specifies the fundamental frequency pattern read from the pitch pattern storage means by the voice quality selection means. Modulation is performed according to the selected voice quality, and the sampling period conversion means determines the sampling period of the discrete voice signal to be D / A converted by the digital / analog conversion means according to the voice quality selected and designated by the voice quality selection means. It is also possible to adopt a configuration in which the second sampling period defined by the above is converted.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of speech analysis, the D / s are converted into the third sampling period different from the second sampling period.
By performing A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices of different voice qualities can be synthesized using the same feature parameter. Of the second sampling period different from the second sampling period and a third sampling period different from the second sampling period, or the fundamental frequency pattern of the second sampling period and synthesis is selected and designated by the voice quality selecting means. The pitch of the voice to be synthesized (voice pitch) can be appropriately controlled by setting the voice quality according to the voice quality, so that a voice with a natural pitch can be synthesized while changing the voice quality.

Next, a speech synthesizer according to a thirteenth aspect of the present invention corresponds to the speech synthesizer according to the fourth aspect, in which the sampling period of the discrete speech signal synthesized by the speech synthesizer is The voice synthesizing unit converts the second sample period to a second sample period different from the first sample period, and the digital / analog converting unit converts the discrete voice signal into the second sample period. It is characterized in that it is configured to be converted into an analog audio signal in a third cycle different from.

In addition, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided. In the voice synthesis means, the sampling period of the synthesized discrete voice signal depends on the voice quality selected and designated by the voice quality selection means. It is also possible to adopt a configuration in which the second sampling period defined by the above is converted.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of creating the speech unit, D is sampled at the third sampling period different from the second sampling period. By performing the / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same voice unit. In order to synthesize the voice with the original voice quality, the first sampling period = the second sampling period = the third sampling period.

Next, a speech synthesizer according to a fourteenth aspect of the present invention corresponds to the speech synthesizer according to the fifth aspect, in which the sampling period of the discrete speech signal synthesized by the speech synthesizer is It is possible to convert the second sampling period different from the first sampling period by the voice synthesizing means, and the sampling process is performed in the synthesizing process when the sampling period conversion to the second sampling period is performed. A speech rate parameter having a value different from that used when the conversion is not performed is used, and the discrete voice signal output from the voice synthesizer is converted by the digital / analog converter to a third sampling period different from the second sampling period. It is characterized in that it is configured to convert into an analog audio signal at a cycle. Here, the speech rate parameter used in the synthesis process may be determined based on the second sampling period and the third sampling period.

In addition to the above, voice quality selecting means for selecting and specifying the voice quality of the voice to be synthesized is further provided, and in the voice synthesizing means, the speech rate parameter to be used and the second sampling period are selected and designated by the voice quality selecting means. It is also possible to adopt a configuration that is determined according to the voice quality.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of creating the speech unit, D is sampled at the third sampling period different from the second sampling period. By performing the A / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same feature parameter. Parameters
When the sample period is not converted (equivalent to converting to the first sample period), it is different from the value used, or is determined according to the second sample period and the third sample period. By determining the second sampling period according to the selected and designated voice quality and determining the speech rate parameter according to the selected and designated voice quality, it is possible to appropriately control the speech rate of the synthesized voice, and thus while changing the voice quality. Speech can be synthesized at a natural speaking rate.

Next, a speech synthesizer according to a fifteenth aspect of the present invention corresponds to the speech synthesizer according to the sixth aspect, wherein the sampling period of the discrete speech signal synthesized by the speech synthesizer is It is possible to convert the second sample period different from the first sample period by the voice synthesizing means, and in the synthesizing process at the time of performing the sample period conversion to the second sample period, the conversion of the sample period is performed. The phoneme duration is determined so that the phoneme duration is different from that when not performing
It is characterized in that the discrete voice signal output from the voice synthesizing means is converted into an analog voice signal by a digital / analog converting means in a third cycle different from the second sampling cycle. Here, the phoneme duration used in the synthesis process may be determined based on the second sampling period and the third sampling period.

In addition to the above, voice quality selecting means for selecting and specifying the voice quality of the voice to be synthesized is further provided, and in the above-mentioned voice synthesizing means, the phoneme duration to be used and the second sampling period are selected and designated by the voice quality selecting means. It is also possible to adopt a configuration that is determined according to the voice quality.

In the present invention, after the synthesized speech is converted into the second sample period different from the first sample period at the time of creating the speech unit, D is sampled at the third sample period different from the second sample period. By performing the A / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same feature parameter. The time is set to be different from the value used when the sampling period is not converted (equivalent to converting to the first sampling period), is set according to the second sampling period and the third sampling period, or the voice quality is set. Since the second sampling period is determined according to the voice quality selected and designated by the selection means and the phoneme duration is determined according to the voice quality selected and designated, the utterance speed of the synthesized voice can be appropriately controlled. change Voice can be synthesized in a natural speech rate while.

Next, a speech synthesizer according to a sixteenth aspect of the present invention corresponds to the speech synthesizer according to the seventh aspect, and is created from a discrete speech signal sampled at the first sampling period. A speech synthesizing unit for synthesizing a discrete speech signal by selecting the selected speech unit based on the given phoneme information and connecting the selected speech units while expanding and contracting in the time axis direction. The speech synthesis means capable of converting the sampling period of the discrete speech signal into a second sampling period different from the first sampling period, and the discrete speech signal output from the speech synthesis means are converted into the second sampling period. And a digital / analog conversion means for converting into an analog audio signal at a third sampling period different from the sampling period. Here, the degree of expansion / contraction in the time axis direction with respect to the voice unit when the voice unit is connected in the voice synthesizing unit may be determined based on the second sampling period and the third sampling period.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of creating the speech unit, D is sampled at the third sampling period different from the second sampling period. By performing the A / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same feature parameter. By connecting while expanding and contracting to, for example, connecting while expanding and contracting in the time axis direction at a degree determined based on the second sampling period and the third sampling period, the utterance speed of the synthesized voice is appropriately adjusted. It is possible to control, and it is also possible to properly control the temporal change of the spectrum transient part of the synthesized voice.
Clear voice can be synthesized at a natural speech rate while changing the voice quality.

In addition to the above, voice quality selection means for selectively designating the voice quality of the voice to be synthesized is further provided, and in the above-mentioned voice synthesis means, the selected voice unit is determined according to the voice quality selected and designated by the voice quality selection means. A configuration in which discrete voice signals are synthesized by expanding and contracting in the time axis direction depending on the degree, and the sampling period of the synthesized discrete voice signals is converted into a second sampling period determined according to the voice quality selected and designated. It is also possible to do so. In this case, the spectrum of the voice is shifted on the frequency logarithmic axis by converting the synthesized voice into the second sample period determined according to the selected voice quality and then performing the D / A conversion in the third sample period. Therefore, it is possible to synthesize voices with different voice qualities by using the same feature parameter.However, when synthesizing, it is necessary to connect the voice units while expanding and contracting in the time axis direction to the degree according to the selected voice quality. By this, the speech rate of the synthesized speech can be appropriately controlled, and the temporal change of the spectrum transient part of the synthesized speech can also be controlled appropriately, so that a clear speech can be synthesized at a natural speech rate while changing the voice quality.

Next, a speech synthesizer according to a seventeenth aspect of the present invention corresponds to the speech synthesizer according to the eighth aspect, in which the discrete speech signal sampled at the first sampling period is A speech unit accumulating unit for accumulating a plurality of speech units cut out in a predetermined synthesis unit from a time series of speech feature parameter frames obtained by analyzing a time window of one frame period, and the speech unit accumulation. Synthesizer parameter frame time series generating means for selecting and connecting speech units based on input phoneme information from the means to generate a time series of the synthesized parameter frame, and the synthesizer parameter frame time series generating means And a synthesizing means for synthesizing the discrete audio signal in the second frame cycle different from the first frame cycle from the time series of Sampling period conversion means for converting the sampling period of the audio signal into a second sampling period different from the first sampling period, and a discrete speech signal whose sampling period has been converted by the sampling period conversion means, the second sampling And a digital / analog conversion means for converting into an analog audio signal at a third sampling period different from the period. Here, the second frame cycle may be determined based on the first frame cycle, the second sampling cycle, and the third sampling cycle.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of creating the speech unit, D is sampled at the third sampling period different from the second sampling period. By performing the A / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same feature parameter. Here, however, the second frame at the time of synthesis is synthesized. By determining the period based on the first frame period, the second sampling period, and the third sampling period, and by making it different from the window period (first frame period) at the time of analysis, Since the speech rate can be appropriately controlled, it is possible to synthesize a voice with a natural speech rate while changing the voice quality.

In addition to the above, voice quality selecting means for selecting and specifying the voice quality of the voice to be synthesized is further provided, and in the voice synthesizing means, the second frame period and the second sample period are selected and designated by the voice quality selecting means. It is also possible to adopt a configuration determined according to the voice quality. In this case, the spectrum of the voice is shifted on the frequency logarithmic axis by converting the synthesized voice into the second sample period determined according to the selected voice quality and then performing the D / A conversion in the third sample period. Therefore, it is possible to synthesize voices having different voice qualities using the same feature parameter. However, here, the second frame period at the time of synthesis is determined according to the selected voice quality, and the window period at the time of analysis ( Since it is possible to appropriately control the utterance speed of the synthesized voice by making it different from the first frame period),
A voice with a natural speaking speed can be synthesized while changing the voice quality.

Next, a speech synthesizer according to an eighteenth aspect of the present invention corresponds to the speech synthesizer according to the ninth aspect, wherein a second sampling period different from the first sampling period and Pitch pattern generation means for generating a fundamental frequency pattern of a voice from input prosody information based on a third sampling period different from the second sampling period; and (a phonological parameter generated based on the input phonological information and the pitch pattern). Sampling period conversion means for converting the sampling period of the synthesized discrete audio signal (from the fundamental frequency pattern generated by the generation means) into the second sampling period, and the sampling period conversion means for converting the sampling period. Digital / analog conversion means for converting the audio signal into an analog audio signal at the third sampling period.

In the present invention, after the synthesized speech is converted into the second sampling period different from the first sampling period at the time of creating the speech unit, D is sampled at the third sampling period different from the second sampling period. By performing the A / A conversion, the spectrum of the voice shifts on the frequency logarithmic axis, so that voices having different voice qualities can be synthesized using the same feature parameter. By determining the basic frequency pattern for synthesis based on the sampling period of 3, the pitch of the voice to be synthesized (voice pitch) can be controlled appropriately, so that a voice with a natural pitch can be synthesized while changing the voice quality.

In addition, voice quality selection means for selecting and designating the voice quality of the voice to be synthesized is further provided, and the pitch pattern generating means uses the prosody information to select the basic voice from the prosody information according to the voice quality selected and designated by the voice quality selection means. A frequency pattern is generated, and the sampling cycle conversion means converts the sampling cycle of the combined discrete speech signal into a second sampling cycle determined according to the voice quality selected and designated by the voice quality selection means. Is also possible. In this case, the spectrum of the voice is shifted on the frequency logarithmic axis by converting the synthesized voice into the second sample period determined according to the selected voice quality and then performing the D / A conversion in the third sample period. Therefore, it is possible to synthesize voices with different voice qualities using the same feature parameter.However, here, by determining the fundamental frequency pattern for synthesis according to the selected voice quality, the pitch of the voices to be synthesized (voice The pitch can be controlled appropriately, so that a voice with a natural pitch can be synthesized while changing the voice quality.

[0074]

Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment] FIG. 1 is a block diagram showing the schematic arrangement of a speech analysis / synthesis apparatus according to the first embodiment of the present invention.

In FIG. 1, 1102 is stored in the memory 11.
For a voice (discrete voice signal) sampled (sampled) at 5 Hz (first sampling period), a frame period (first sampling period)
Frame period), a Hanning window with a window width of 20 msec is applied at 10 msec, and the 0th to 2nd order obtained by the procedure described in the related art.
A time series (feature parameters) of parameter frames of low-order cepstrum coefficients up to the fifth order and voiced / unvoiced information of voice corresponding to each frame are stored. The low-order cepstrum coefficient has already been described in the prior art, and will be omitted here.

On the other hand, the memory 12 stores a time-series pattern of the fundamental frequency of the voice, which is obtained from the same voice by the predetermined fundamental frequency extraction method.

The synthesis filter processing unit 13 reads out each data from these two memories 11 and 12, and
An LMA filter (not shown) having a filter coefficient of the parameter frame (cepstral coefficient) of the voice read out is used as a time series pattern (pitch pattern) of the fundamental frequency of the voice read from the memory 12 in the voiced section. A desired voice (discrete voice signal) is synthesized by driving with random noise in the unvoiced section with a periodic pulse based on the above.

Since the processing up to this point is performed by the program, the sound output from the synthesis filter processing unit 13 (the LMA filter therein) is a discrete sound signal. Therefore, this discrete audio signal is supplied to the D / A converter 14 and converted into an electrical analog signal. The audio analog signal thus obtained is amplified by the amplifier 15, and the speaker 16
By driving, it is possible to obtain audio that can be perceptually heard.

The process up to this point is almost the same as the example (of the analyzing and synthesizing apparatus of FIG. 23) described in the prior art. The point of this embodiment is that a voice quality switching unit 17 and a voice quality control unit 18 are added. Although the D / A converter 14 is composed of hardware, the sampling frequency for the conversion can be controlled by software.

The voice quality switching section 17 is capable of switching and designating the voice quality at the time of synthesis by a user or an application program. In the present embodiment, it is assumed that the voice quality switching unit 17 can specify three types of voice qualities.

The voice quality control unit 18 sets 11025 Hz, 12000 HZ, according to the voice quality designated by the voice quality switching unit 17.
Any sampling frequency of 10000 Hz (second
The D / A converter 14 is controlled so as to perform digital / analog conversion (D / A conversion) at the sampling period of.

The same sampling frequency as the voice sampling frequency (first sampling period) when the cepstrum stored in the memory 11 is created, that is, 11025 Hz.
Then, if the D / A converter 14 performs D / A conversion, it is possible to perform voice synthesis with the voice quality of the original voice.

On the other hand, the D / A converter 14 uses another sampling frequency (second sampling period different from the first sampling period).
As shown in FIG. 2, if the D / A conversion is performed, the effect of shifting the sound spectrum (spectrum envelope in the figure) in the frequency axis direction is obtained, so that the individuality of the sound is changed and the analog sound signal thus obtained is obtained. Has a voice quality different from that of the original voice.

[Second Embodiment] In the first embodiment, by providing the voice quality switching unit 17 and the voice quality control unit 18, the voice quality of the synthesized voice can be easily increased, but it is synthesized. Voice speed varies depending on voice quality.

That is, when (sampling frequency when creating the cepstrum)> (sampling frequency for D / A conversion), the speed of the synthesized voice becomes slow. vice versa,
(Sampling frequency when creating the cepstrum) <(D /
At the sampling frequency of A conversion), the speed of the synthesized voice is high.

The difference in the speed of the synthesized voice does not cause much problem if the sampling frequencies at the time of D / A conversion are different from those in the first embodiment (91%, 109%). However, as described below, the sampling frequency (the second sampling period) at the time of D / A conversion is significantly different from the sampling frequency (the first sampling period) of the voice when the cepstrum stored in the memory 11 is created. In some cases, it becomes a problem.

First, if the ratio of the sampling frequency at the time of creating the cepstrum and the sampling frequency of the D / A conversion at the time of synthesis is close to 1, the voice quality changes little. Conversely, if the ratio deviates from 1, the voice quality changes greatly. Therefore, if the voice quality is to be changed significantly, the ratio of these two sampling frequencies may be set to, for example, about 50% and 200%. With this, the speed of the synthesized voice is 50% and 200% of the original voice, respectively. In other words, it will be doubled, and it will be hard to hear.

Therefore, a second embodiment will be described in which the speed of synthesized speech can be made constant even when the sampling frequency for D / A conversion during synthesis is significantly different from the sampling frequency for creating the cepstrum.

FIG. 3 is a block diagram showing a schematic configuration of a voice analysis / synthesis apparatus according to the second embodiment of the present invention. The same parts as those in FIG. 1 are designated by the same reference numerals.

The point of the present embodiment is that, instead of the synthesizing filter processing unit 13 in FIG. 1, a synthesizing filter processing unit 23 capable of controlling the frame period at the time of synthesizing is provided, and
Instead of the voice quality control unit 18 in the middle, a voice quality control unit 28 for controlling not only the sampling frequency of D / A conversion but also the frame period at the time of synthesis is provided, and the voice quality control unit 28 responds to the designation of the voice quality switching unit 17. D / A converter 14
At the same time as the sampling frequency of the A conversion, the frame period at the time of synthesis in the synthesis filter processing unit 23 is controlled.

In this embodiment, three types of voice qualities can be designated by the voice quality switching section 17 as in the first embodiment. The voice quality control unit 28 determines, for example, 11025 Hz, 8000 Hz, 16 according to the voice quality designated by the voice quality switching unit 17.
The D / A converter 14 is controlled to perform D / A conversion at any sampling frequency of 000 Hz.

At the same time, the voice quality control unit 28 uses the voice quality switching unit 17
The frame period of the synthesis performed by the synthesis filter processing unit 23 is set according to the voice quality designated by. As a result, in the synthesis filter processing unit 23, the parameter frame (cepstral coefficient) of the voice read from the memory 11 is set at the set frame period (to the LMA filter).
Input and synthesize speech (discrete speech signal) at the frame cycle.

The frame period of synthesis is given by the following equation.

(Frame period) = (Analysis frame period) × (Analysis sampling period) / (D / A conversion sampling period) = (Analysis frame period) × (D / A conversion sampling frequency) / (Analysis sampling) Therefore, when performing D / A conversion at the same sampling frequency as the sampling frequency (first sampling period) of the voice at the time of creating the cepstrum (at the time of analysis), that is, at 11025 Hz, the voice quality control unit 28 uses the above equation. The synthesis filter processing unit 13 is controlled so that the synthesis frame period is synthesized in 10 msec, which is the same as the analysis frame period (first frame period) which is the frame period when the cepstrum is created. However,
It is assumed that the cepstrum stored in the memory 11 is created under the same conditions as in the first embodiment.

Further, the voice quality control section 28 sets the D / A at a sampling frequency (second sampling period different from the first sampling period) different from the sampling frequency of the voice when the cepstrum is created (at the time of analysis), for example, 8000 Hz. When converting, 10 [msec] x 8000 [Hz] / 11025 [Hz] = 7.3 [m]
sec] frame period (second frame period different from the first frame period) is controlled to perform synthesis, and 16000H
When performing D / A conversion in z, 10 [msec] × 16000 [Hz] / 11025 [Hz] = 14.
The control is performed so that the composition is performed in a frame period of 5 [msec] (a second frame period different from the first frame period).

As described above, in this embodiment, the change in the speed of the synthesized voice when the D / A conversion is performed at the sampling frequency different from the analysis time (the second sampling period different from the first sampling period) is synthesized. The frame period (second frame period) of the synthesis in the filter processing unit 23 is the frame period (first frame period) when the cepstrum is created (at the time of analysis).
It can be offset by making it different from.

Therefore, D / A at 8000 Hz or 16000 Hz, which is different from the sampling frequency at the time of analysis,
Even if the conversion is performed, it is possible to obtain the analog signal of the voice having the same speed.

[Third Embodiment] In the first embodiment, when the sampling frequency of D / A conversion is changed from that at the time of analysis, the pitch of voice, that is, the pitch of voice is changed. .

That is, when (sampling frequency when creating the cepstrum)> (sampling frequency for D / A conversion), the pitch of the synthesized voice becomes low. vice versa,
(Sampling frequency when creating the cepstrum) <(D /
At the sampling frequency of A conversion), the pitch of synthesized speech becomes high.

The difference in the pitch of the synthesized voices does not cause much problem if the sampling frequencies at the time of D / A conversion are different from those in the first embodiment (91%, 109%). However, in an attempt to significantly change the voice quality,
If the ratio of both sampling frequencies is set to, for example, about 50% and 200%, the pitch of the synthesized voice is 50%,
Since it changes to 200%, it is the same as when creating the cepstrum. 1
Compared with the voice (or the original voice) when D / A converted at 1025 Hz, the former synthesizes a voice with a pitch lower by 1 [oct] (octave), and the latter synthesizes a voice with a higher pitch of 1 [oct]. The problem that it becomes difficult to hear occurs.

Therefore, a third embodiment will be described in which the pitch of the synthesized voice can be made constant even when the sampling frequency of the D / A conversion at the time of synthesis is significantly different from the sampling frequency at the time of creating the cepstrum.

FIG. 4 is a block diagram showing the schematic arrangement of a speech analysis / synthesis apparatus according to the third embodiment of the present invention. The same parts as those in FIG. 1 are designated by the same reference numerals.

The point of this embodiment is to replace the synthesis filter processing unit 13 in FIG. 1 with a synthesis filter processing unit 33 capable of controlling the frame period at the time of synthesis, and to use the memory 12 and the synthesis filter processing unit 33. A pitch modulation processing unit for converting (pitch modulation) a basic frequency pattern (time series) (pitch pattern) read from the memory 12 into another basic frequency pattern having a different frequency and giving it to the synthesis filter processing unit 33. 31 is provided, and instead of the voice quality control unit 18 in FIG. 1, a voice quality control unit 38 is provided for controlling not only the sampling frequency of D / A conversion but also the frame period and pitch modulation at the time of synthesis. According to the designation of the voice quality switching unit 17, the D / A converter 14
The sampling frequency and the frame period at the time of synthesis in the synthesis filter processing unit 33 are controlled, and at the same time, the pitch modulation in the pitch modulation processing unit 31 is controlled.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 17 as in the first embodiment. The voice quality control unit 38 determines, for example, 11025 Hz, 8000 Hz, 16 according to the voice quality designated by the voice quality switching unit 17.
The D / A converter 14 is controlled to perform D / A conversion at any sampling frequency of 000 Hz.

The voice quality control unit 38 controls the D / A converter 14 according to the voice quality designated by the voice quality switching unit 17.
At the same time as setting the sampling frequency for conversion, the frame period for combining performed by the combining filter processing unit 33 is set. The frame period of synthesis is given by the following equation.

(Frame period) = (Analysis frame period) × (Analysis sampling period) / (D / A conversion sampling period) = (Analysis frame period) × (D / A conversion sampling frequency) / (Analysis sampling) Frequency) The cepstrum stored in the memory 11 is the first
It is assumed that it is created under the same conditions as in the embodiment.

The voice quality control section 38 further determines that the pitch (pitch pattern) given to the synthesis filter processing section 33 is (pitch given to the synthesis filter processing section 33) = (pitch read from the memory 12) × (D / A conversion) Sampling period) / (sampling period during analysis) = (pitch read from memory 12) × (sampling frequency during analysis) / (sampling frequency for D / A conversion) Control.

Therefore, when performing D / A conversion at the same sampling frequency as the sampling frequency (first sampling period) of the voice at the time of creating the cepstrum, that is, 11025 Hz, the voice quality control section 38 uses the above-mentioned expression to synthesize the filter. The pitch modulation processing unit 31 is controlled so that the pitch given to the processing unit 33 becomes the same as that at the time of analysis.

Further, the voice quality control section 38 performs D / A at a sampling frequency (second sampling period different from the first sampling period) different from the sampling frequency of the voice when the cepstrum is created (at the time of analysis), for example, 8000 Hz. When converting, set the pitch during analysis to (11025 [Hz] / 8000
[Hz]) so that it is given to the synthesis filter processing unit 33,
When performing D / A conversion at 16000 Hz, the pitch modulation processing unit 31 is controlled so as to multiply the analysis pitch by (11025 [Hz] / 16000 [Hz]) and give it to the synthesis filter processing unit 33. .

As described above, in the present embodiment, the pitch given to the synthesis filter processing unit 33 is previously modulated by the pitch modulation processing unit 31 under the control of the voice quality control unit 38, which is different from the time of analysis. D / at sampling frequency
It is possible to cancel the change in the pitch of the synthetic voice that occurs when the A conversion is performed.

Therefore, D / A at 8000 Hz or 16000 Hz different from the sampling frequency at the time of analysis
Even if the conversion is performed, it is possible to obtain an analog signal of voice having the same pitch.

[Fourth Embodiment] FIG. 5 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the fourth embodiment of the present invention.

This speech rule synthesizing device is realized by executing dedicated software (sentence / speech conversion software) on an information processing device such as a personal computer, and has a sentence / speech conversion (TTS) processing function, that is, It has a sentence-speech conversion processing (sentence-speech synthesis processing) function for generating speech from a text, and the functional configuration thereof is roughly divided into a language processing unit 41 and a speech synthesis unit 42.

The language processing unit 41 analyzes the input sentence, for example, a sentence containing kanji and kana to generate reading information and accent information, and a phonetic symbol string in which phonological symbol sequence / accent information is described based on the information. Manages the process of generating.

The voice synthesizing unit 42 controls the process of generating a voice based on the voice symbol string output from the language processing unit 41.

Now, in the speech rule synthesizing device of FIG.
A document (Japanese document in this case) to be subjected to sentence-to-speech conversion (speech) is stored as a text file 43. According to the sentence / speech conversion software, this device reads out the sentence containing kanji and kana from the file 43 one by one, and the language processing unit 41 and the speech synthesis unit 422 perform the sentence / speech conversion processing described below to synthesize the speech.

First, the kanji / kana mixed sentence read from the text file 43 is input to the language analysis processing unit 44 in the language processing unit 41.

The language analysis processing unit 44 performs morphological analysis on the input kanji / kana mixed sentence to generate reading information and accent information. Morphological analysis is an operation of analyzing which character string forms a phrase in a given sentence, and what the structure of the word is.

For this purpose, the language analysis processing unit 44 uses a morpheme dictionary 45 having a morpheme, which is the minimum constituent element of a sentence, as an entry word, and a connection rule file 46 in which connection rules between morphemes are registered. That is, the language analysis processing unit 44
All the morpheme sequence candidates obtained by matching the input sentence with the morpheme dictionary 45 are obtained (brute force method), and from among them, the connection rule file 46 is referred to and the combinations that can be connected grammatically before and after are output. . In the morpheme dictionary 45,
Along with grammatical information used at the time of analysis, morpheme readings and accent types are registered. For this reason, if a morpheme is determined by morphological analysis, reading and accent type can be given at the same time.

For example, when the morphological analysis is performed on the sentence "go to the park and read a book", / park / f / go / book / of / read / read /.

Is divided into morphemes. Reading and accent type are given to each morpheme, and it becomes / Kouen / E / Itte / Hon / wo / Yomi / Mas /. Here, a morpheme containing "^" means that the pitch is high in the syllable immediately before and the pitch is low in the syllable immediately after. When there is no "^", it means that the accent is flat.

By the way, when humans read a sentence, they do not read by adding accents in units of such morphemes, but by reading several morphemes together and adding accents to the unit.

Therefore, in consideration of such a situation, the language analysis processing unit 44 further collects morphemes by one accent phrase (unit giving an accent) and at the same time estimates movement of the accent due to the combination. In addition to this, the language analysis processing unit 44 also adds information such as vowel devoicing and pause (breathing) during reading, and in the above example, finally generates the following reading information.

/ Cohenye / Itte. / Ho ^ o / Yomima ^ (s) / Here, the period "." Represents breath and "()" represents a syllable in which a vowel is devoiced.

Now, when the reading information is generated by the language analysis processing unit 44 in the language processing unit 41 as described above,
The phoneme duration calculation processing unit 47 in the voice synthesis unit 42 is activated. The phoneme duration calculation processing unit 47, according to the reading information generated by the language analysis processing unit 44, the duration time (unit is ms) of the consonant part and the vowel part of each syllable included in the input sentence.
To determine.

The process of determining the duration in the phoneme duration calculation unit 47 is performed by the boundary (C) between the consonant (C) and the vowel (V).
This is realized by an extremely simple algorithm in which the positions of (V crossing) are arranged at equal intervals.

The CV crossover interval (CV crossover interval as a speech rate parameter) is given by the speech rate control section 48 in the voice synthesis section 42. Although not shown, the software used in this embodiment allows the user to specify the speed of synthesized speech. And
By giving the speed of the voice designated by the user to the speech rate control unit 48, the speech rate control unit 48 concerned
Is adjusting the interval of the CV crossing (determined by the duration determining process in the phoneme duration calculating unit 47) to actually change the speed of the synthesized voice. However, it is known from the analysis results that the duration of consonants in Japanese speech is almost constant even if the utterance speed is changed, so the duration of consonants is kept constant and the duration of vowels is adjusted. C
Change the interval between V crossings.

When the phoneme duration calculation processing section 47 determines the duration of each syllable (consonant section and vowel section) included in the input sentence, the pitch generation processing section 49 in the same speech synthesis section 42 is activated. It The pitch generation processing unit 49 determines the duration determined by the phoneme duration calculation processing unit 47,
First, the point pitch position is set based on the accent information determined by the language analysis processing unit 44 (in the language processing unit 41). Next, the pitch generation processing unit 49 interpolates a plurality of set point pitches with a straight line to obtain a pitch pattern (fundamental frequency pattern) every 10 msec, for example.

On the other hand, the phonological parameter generation processing unit 50 in the speech synthesis unit 42 generates phonological parameters based on the phonological information of the speech symbol string passed from the language analysis processing unit 44 (in the language processing unit 41). The processing is performed as follows in parallel with the pitch pattern generation processing by the pitch generation processing unit 49, for example.

First, in this embodiment, an actual voice sampled at a sampling frequency of 11025 Hz (first sampling period) was obtained by analysis with a modified cepstrum method at a window length of 20 msec and a frame period (first frame period) of 10 msec. A speech unit file in which a total of 137 speech units, which are obtained by cutting out all syllables necessary for synthesizing Japanese speech from 0th to 25th order cepstral coefficients in units of consonant + vowel (CV) (not shown) ) Is prepared. The contents of this speech unit file are read into a speech unit area (hereinafter referred to as a speech unit memory) 51 secured in, for example, a main memory (not shown) at the start of the sentence-speech conversion process according to the sentence-speech conversion software. It is assumed that

The phoneme parameter generation processing unit 50 stores the above-mentioned CV-based speech units in the speech unit memory 51 in accordance with the phoneme information in the speech symbol string passed from the language analysis processing unit 44 (in the language processing unit 41). Are sequentially read, and the phoneme parameters (feature parameters) of the speech to be synthesized are generated by connecting the read speech units.

When the pitch generation processing unit 49 generates a pitch pattern and the phoneme parameter generation processing unit 50 generates phoneme parameters, the synthesis filter processing unit 52 in the speech synthesis unit 42 is activated. As shown in FIG. 6, the synthesis filter processing unit 52 includes a white noise generating unit 521, an impulse generating unit 522, and a driving sound source switching unit 52.
3 and an LMA filter 524, and synthesizes speech from the generated pitch pattern and phonological parameters as follows.

First, in the voiced part (V) of the voice, the driving sound source switching part 523 switches it to the impulse generating part 522 side. The impulse generation unit 522 generates impulses at intervals according to the pitch pattern generated by the pitch generation processing unit 49, and uses this impulse as a sound source for the LM.
The A filter 524 is driven.

On the other hand, in the unvoiced part (U) of the voice, the driving sound source switching part 523 switches it to the white noise generating part 521 side. The white noise generator 521 generates white noise, and the white noise is used as a sound source for the LM.
The A filter 524 is driven. The LMA filter 524 uses the sound cepstrum directly as the filter coefficient.

Since the phonological parameter generated by the phonological parameter generation processing unit 50 in this embodiment is the cepstrum as described above, this phonological parameter becomes the filter coefficient of the LMA filter 524 and is switched by the driving sound source switching unit 523. When driven by a sound source, it outputs synthetic speech.

The voice synthesized by the synthesis filter processing unit 52 (the LMA filter 524 therein) is a discrete voice signal, which is converted into an analog signal by the D / A converter 53 and output to the speaker 55 through the amplifier 54. For the first time, it can be heard as a sound.

The processing up to this point is almost the same as the example described in the "Prior Art" section with reference to FIG.

The point of this embodiment is that the voice synthesis unit 42
A voice quality switching unit 56 and a voice quality control unit 57 (corresponding to the voice quality switching unit 17 and the voice quality control unit 18 in FIG. 1) are added therein. The D / A converter 53 is composed of hardware, but the sampling frequency for the conversion can be controlled by software.

The voice quality switching section 56 is capable of switching the voice quality at the time of synthesis by a user designation or an application program. In this embodiment, it is assumed that three types of voice qualities can be designated by the voice quality switching unit 56.

The voice quality control unit 57 operates at 11025 Hz, 12000 HZ, depending on the voice quality designated by the voice quality switching unit 56.
D / at any sampling frequency of 10000 Hz
The D / A converter 53 is controlled to perform A conversion.

If the D / A converter 53 performs D / A conversion at the same sampling frequency (first sampling period) as when the speech unit was created, that is, 11025 Hz, the voice quality of the original voice Can be synthesized. On the other hand, the D / A converter 53 has another sampling frequency (first
If the D / A conversion is performed at a second sampling period (which is different from the sampling period), the effect of shifting the voice spectrum in the frequency axis direction as shown in FIG. 2 is obtained as already described in the first embodiment. As a result, the individuality of the voice changes, and the voice quality of the analog voice signal thus obtained differs from the voice quality of the voice that is the source of the voice segment.

[Fifth Embodiment] In the fourth embodiment, since the voice quality switching unit 56 and the voice quality control unit 57 are provided, the voice quality of the synthesized voice can be easily increased, but the voice is synthesized. Voice speed varies depending on voice quality.

That is, when (sampling frequency when creating a segment)> (sampling frequency for D / A conversion), the speed of the synthesized voice becomes slow. On the other hand, when (sampling frequency at the time of creating a segment) <(sampling frequency for D / A conversion), the speed of the synthesized voice becomes faster.

The difference in the speed of the synthesized voice does not cause much problem if the sampling frequencies at the time of D / A conversion are the same as those in the fourth embodiment (91%, 109%). However, as will be described below, when the sampling frequency at the time of D / A conversion is significantly different from the sampling frequency of the voice when the voice unit read in the voice unit memory 51 is significantly different, there is a problem.

First, if the ratio of the sampling frequency at the time of creating a segment and the sampling frequency of D / A conversion at the time of synthesis is close to 1, the voice quality changes little, and conversely if the ratio deviates from 1, the voice quality changes greatly. . Therefore, if the voice quality is to be changed significantly, the ratio of these two sampling frequencies may be set to, for example, about 50% and 200%, but with this, the speed of the synthesized voice is 50% and 2% of the original voice, respectively.
00%, that is, half the time, making it difficult to hear.

Therefore, a fifth embodiment will be described in which the speed of the synthesized speech can be made constant even when the sampling frequency of the D / A conversion at the time of synthesis is significantly different from the sampling frequency at the time of creating the segment.

FIG. 7 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the fifth embodiment of the present invention. The same parts as those in FIG. 5 are designated by the same reference numerals.

The point of the present embodiment is to provide a speech rate control section 68 capable of controlling the speed (speech rate) of synthesized speech instead of the speech rate control section 48 in FIG.
In place of the voice quality control unit 57 in FIG. 5, a voice quality control unit 67 for controlling not only the sampling frequency of D / A conversion but also the speed of synthesized speech is provided, and the voice quality control unit 67 specifies the voice quality switching unit 56. Accordingly, D in the D / A converter 53
This is to control the speed of the synthesized voice in the speech rate control unit 48 at the same time as the sampling frequency of the / A conversion.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 56 as in the fourth embodiment. The voice quality control unit 67 determines, for example, 11025 Hz, 8000 Hz, 16 according to the voice quality designated by the voice quality switching unit 56.
The D / A converter 53 is controlled to perform D / A conversion at any sampling frequency of 000 Hz.

At the same time, the voice quality control unit 67 controls the voice quality switching unit 56.
According to the voice quality designated by
Is controlled as follows. That is, the voice quality control unit 67 determines that the CV crossover interval described above is (CV crossover interval) = (CV crossover interval during D / A conversion at 11025 Hz) × (D / A conversion sampling frequency) / ( The speech rate control unit 68 is controlled so that it becomes the sampling frequency when the segment is created).

Therefore, when the D / A conversion is performed at 8000 Hz, the voice quality control unit 67 calculates
The CV crossing interval (1 for D / A conversion at 25 Hz
The utterance speed control unit 68 is controlled so that the CV crossing interval is 1025 [Hz] / 8000 [Hz] times. Further, the voice quality control unit 67, when performing D / A conversion at 16000 Hz,
The utterance speed control unit 68 is controlled so that the CV spread interval is (11025 [Hz] / 16000 [Hz]) times the CV spread interval when D / A conversion is performed at 11025 Hz.

As described above, in this embodiment, the change in the speed of the synthesized voice when the D / A conversion is performed at the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of creating the speech unit. Can be canceled by changing the interval between CVs (speech rate parameter).
Therefore, even if the D / A conversion is performed at 8000 Hz or 16000 Hz, which is different from the sampling frequency at the time of creating the voice unit, it is possible to obtain a voice analog signal having substantially the same speed.

[Sixth Embodiment] If rule synthesis is performed based on the fifth embodiment, it is possible to easily realize that the speed of synthesized speech can be kept substantially constant while changing the voice quality. However, as described above, in the process (phoneme duration calculation process) in the phoneme duration calculation processing unit 47, the phoneme duration of the consonant is made constant for each syllable even if the interval between CV crossings is changed. If the D / A conversion is performed at a sampling frequency different from that at the time of creating one piece, the duration of the consonant may be shortened or extended, and as a result, the clarity and naturalness of the synthesized voice may be affected.

Therefore, a sixth embodiment will be described in which the duration of the consonant of the synthesized voice can be kept constant for each syllable.

FIG. 8 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the sixth embodiment of the present invention. The same parts as those in FIG. 5 or 7 are designated by the same reference numerals.

The point of this embodiment is that, instead of the phoneme duration calculation processing section 47 in FIG. 7, a phoneme duration calculation processing section 77 whose phoneme duration is controllable is provided.
Instead of the voice quality control unit 67 in the middle, a voice quality control unit 87 for controlling the speed of the synthesized voice according to the phoneme duration as well as the sampling frequency of the D / A conversion is provided, and the voice quality control unit 87 controls the voice quality switching unit 56. According to the designation, the phoneme duration in the phoneme duration calculation processing unit 77 is controlled at the same time as the sampling frequency of the D / A conversion in the D / A converter 53.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 56 as in the fifth embodiment. The voice quality control unit 87 determines, for example, 11025 Hz, 8000 Hz, 16 according to the voice quality designated by the voice quality switching unit 56.
The D / A converter 53 is controlled to perform D / A conversion at any sampling frequency of 000 Hz.

At the same time, the voice quality control section 87 controls the voice quality switching section 56.
The phoneme duration calculation processing unit 77 is controlled as follows according to the voice quality designated by. That is, the voice quality control unit 87
Is the duration of all phonemes, that is, the duration of consonants and the duration of vowels (phoneme duration) = (phoneme duration during D / A conversion at 11025 Hz) × (sampling frequency of D / A conversion)
The phoneme duration calculation processing unit 77 is controlled so that it becomes / (sampling frequency at the time of segment production).

Therefore, when the D / A conversion is performed at 8000 Hz, the voice quality control unit 87 calculates 110
Of the phoneme duration (80 for D / A conversion at 25 Hz)
The phoneme duration calculation processing unit 77 is controlled so that the phoneme duration becomes 00 [Hz] / 11025 [Hz]) times. When performing D / A conversion at 16000 Hz, the voice quality control unit 87 has a phoneme duration that is (16000 [Hz] / 11025 [Hz]) times the phoneme duration when D / A conversion is performed at 11025 Hz. The phoneme duration calculation processing unit 77 is controlled as follows.

As described above, in this embodiment, the change in the speed of the synthesized voice when the D / A conversion is performed at the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of creating the speech unit. Can be canceled by changing each phoneme duration, and the duration of the consonant of the synthesized voice can be kept constant for each syllable.

[Seventh Embodiment] If synthesis is performed based on the fifth or sixth embodiment, the speed of synthesized speech can be kept substantially constant while changing the voice quality. However, the difference between the sampling frequency at the time of creating a voice unit and the sampling frequency at the time of D / A conversion is almost the same as the fast-forwarding or slow-moving of a record. It is unavoidable to extend it.

To give an example, the vocalization "/ wa /" is performed by the vocal tract, which causes the mouth to open suddenly and move to / a / from the shape of the mouth close to / u /. Therefore, if the sampling frequency at the time of D / A conversion is lowered and the record is delayed, this change becomes gentle, and not / wa /
It sounds like / uu /.

Therefore, a seventh embodiment will be described in which it is possible to suppress the temporal shrinkage or extension of the synthesized voice transient portion that occurs when the D / A conversion is performed at a sampling frequency different from that at the time of creating the segment. .

FIG. 9 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the seventh embodiment of the present invention. The same parts as those in FIG. 8 are designated by the same reference numerals.

The point of the present embodiment is that, instead of the phoneme parameter generation processing unit 50 in FIG. 8, a phoneme parameter generation processing unit to which a function of expanding and contracting a phoneme parameter (consisting of speech units) in the time axis direction is added. With 90
Instead of the voice quality control unit 87 in FIG. 8, a voice quality control unit 97 is provided for controlling not only the sampling frequency and phoneme duration of D / A conversion but also the expansion and contraction of the phoneme parameters in the time axis direction. Thus, in addition to controlling the sampling frequency of D / A conversion in the D / A converter 53 and the phoneme duration in the phoneme duration calculation processing unit 77 according to the designation of the voice quality switching unit 56, the phoneme parameter generation processing is performed. Part 90
When the transient part of the synthetic speech is shortened, the phoneme parameter is stretched in advance in the time direction to create the phoneme parameter. The parameter is compressed in the time direction to create a phoneme parameter.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 56 as in the sixth embodiment. The voice quality control unit 97 determines 11025 Hz, 8000 Hz, 16000 according to the voice quality designated by the voice quality switching unit 56.
The D / A converter 413 is controlled to perform D / A conversion at any sampling frequency of Hz.

At the same time, the voice quality control unit 97 controls the voice quality switching unit 56.
Depending on the voice quality specified by, the duration of all phonemes, that is, the duration of consonants and the duration of vowels is (phoneme duration) = (phoneme duration at D / A conversion at 11025 Hz) × (D / A conversion sampling frequency)
The phoneme duration calculation processing unit 77 is controlled so that it becomes / (sampling frequency at the time of segment production).

Further, the voice quality control unit 97 controls the phonological parameter generation processing unit 90, and sets the sampling frequency of D / A conversion (second sampling period) to the sampling frequency (first sampling period) at the time of segment production. If the transitional part of the synthetic speech is shortened by making different, the phoneme parameters are stretched in advance in the time direction to create the phoneme parameters. A phonological parameter is created by compressing the phonological parameter in the time direction.

More precisely, the voice quality control unit 97 expands or contracts the length of the segment itself by (sampling frequency of D / A conversion / sampling frequency when the segment is created), and then interpolates the connection. , Generate phonological parameters.

That is, the voice quality control unit 97 in this embodiment.
When D / A conversion is performed at 11025 Hz, the phoneme parameter is generated without expanding or contracting the speech unit, and
When performing D / A conversion at 00Hz, the speech unit is set to (8
000 [Hz] / 11025 [Hz]) times the length, and then connect and interpolate to generate phoneme parameters, and when D / A conversion is performed at 16000 Hz, the speech unit is (16000 [H]
z] / 11025 [Hz]) times longer, and then the phoneme parameter generation processing unit 90 is controlled so as to generate a phoneme parameter by connection interpolation.

As described above, in the present embodiment, the time of the synthetic speech transient portion which occurs when the D / A conversion is performed at the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of creating the segment. The contraction and the extension in the direction can be canceled by expanding and contracting the speech unit in advance when the phoneme parameters are generated.

[Eighth Embodiment] In the fifth to seventh embodiments described above, only voice rule synthesis using a cepstrum or LPC, that is, voice rule synthesis using parameters obtained by analyzing a voice waveform is performed. Instead, it can be applied to waveform synthesis (rule synthesis by). However, in the voice rule synthesis using the parameters, the speed of the synthesized voice can be kept constant while the voice quality is changed, and the contraction / extension of the voice transition part does not occur without using the fifth to seventh embodiments. Various methods are applicable.

Therefore, an explanation will be given of an eighth embodiment in which this simple method is applied to voice rule synthesis using parameters.

FIG. 10 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the eighth embodiment of the present invention.
The same parts as those in FIG. 5 are designated by the same reference numerals.

The point of this embodiment is that, instead of the synthesis filter processing unit 52 in FIG. 5, a synthesis filter processing unit 112 capable of controlling the frame period at the time of synthesis is provided, and the voice quality control unit 57 in FIG. Instead, a voice quality control unit 117 that controls not only the sampling frequency of D / A conversion but also the frame period at the time of synthesis is provided, and according to the designation of the voice quality switching unit 56, the D / A converter is provided by the voice quality control unit 117. 53
At the same time as the sampling frequency of the D / A conversion in (1), the frame period at the time of synthesis in the synthesis filter processing unit 112 is controlled.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 56 as in the fourth embodiment. The voice quality control unit 117 determines 11025 Hz, 8000 Hz, 1600 according to the voice quality designated by the voice quality switching unit 56.
The D / A converter 53 is controlled to perform D / A conversion at any sampling frequency of 0 Hz.

At the same time, the voice quality control unit 117 causes the voice quality switching unit 5 to
According to the voice quality designated by No. 6, the frame period of the synthesis performed by the synthesis filter processing unit 112 is set. As a result, the synthesis filter processing unit 112 inputs the phoneme parameters (cepstrum) generated by the phoneme parameter generation processing unit 50 at the set frame cycle (to the LMA filter), and outputs the voice (discrete voice signal) at the frame cycle. To synthesize.

The frame period of synthesis is given by the following equation.

(Frame Cycle) = (Frame Cycle During Element Creation) x (Sampling Cycle During Element Creation) / (D / A Conversion Sampling Cycle) = (Frame Cycle During Element Creation) x (D / Sampling frequency of A conversion) / (Sampling frequency when creating a voice segment) Therefore, D / A conversion is performed at the same sampling frequency as the voice sampling frequency (first sampling period) when creating a voice voice unit, that is, 11025 Hz. At this time, the voice quality control unit 117 controls the synthesis filter processing unit 112 based on the above equation so that the synthesis frame period is synthesized at 10 msec which is the same as the frame period (first frame period) at the time of creating the speech unit. .

Further, the voice quality control unit 117 performs D / A conversion at a sampling frequency (second sampling period different from the first sampling period) different from the sampling frequency of the voice when the speech unit is created, for example, 8000 Hz. When performing, 10 [msec] × 8000 [Hz] / 11025 [Hz] = 7.3 [m
sec] frame period (second frame period different from the first frame period) is controlled to perform synthesis, and 16000H
When performing D / A conversion in z, 10 [msec] × 16000 [Hz] / 11025 [Hz] = 14.
The control is performed so that the composition is performed in a frame period of 5 [msec] (a second frame period different from the first frame period).

As described above, in this embodiment, the change in the speed of the synthesized voice when the D / A conversion is performed at the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of creating the speech unit. The synthesis filter processing unit 112
This can be offset by making the frame cycle of synthesis (2nd frame cycle) in step 2 different from the frame cycle (first frame cycle) at the time of creating the speech unit.

Therefore, 8000 Hz or 16000 H, which is different from the sampling frequency at the time of creating a speech unit
Even if D / A conversion is performed with z, it is possible to obtain an analog signal of voice with the same speed. At the same time, it is possible to prevent the contraction or extension of the voice transient portion which occurs when the D / A conversion is performed at a sampling frequency different from that at the time of creating the voice unit.

[Ninth Embodiment] There is another problem in the fourth to eighth embodiments. It ’s D /
This means that if the sampling frequency of A conversion is changed to the one used when the segment was created, the pitch of the voice, that is, the pitch of the voice changes. For example, when (sampling frequency when creating a segment)> (sampling frequency for D / A conversion), the pitch of synthesized speech is low. On the other hand, when (sampling frequency at the time of segment production) <(sampling frequency for D / A conversion), the pitch of the synthesized voice becomes high.

The difference in the pitch of the synthesized voices does not cause much problem if the sampling frequencies at the time of D / A conversion are different from those in the fourth embodiment (91%, 109%).

However, if the ratio of both sampling frequencies is set to, for example, about 50% and 200% in order to largely change the voice quality, the pitches of the synthesized voices are 50% and 2 respectively.
It will be 00%. In this case, compared to the voice when D / A converted at 11025 Hz, the former is synthesized with a voice having a pitch lower by 1 [oct], and the latter is synthesized with a voice having a pitch higher than 1 [oct], which makes it difficult to hear. Occurs.

Therefore, a ninth embodiment will be described in which the pitch of synthesized speech can be made constant even if the sampling frequency of D / A conversion at the time of synthesis is significantly different from the sampling frequency at the time of speech unit creation. .

FIG. 11 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the ninth embodiment of the present invention.
The same parts as those in FIG. 5 are designated by the same reference numerals.

The point of this embodiment is that, instead of the synthesis filter processing unit 52 in FIG. 5, the synthesis filter processing unit 13 capable of controlling the frame period and the pitch of synthesized speech during synthesis.
2 is provided, and the pitch pattern (fundamental frequency pattern) generated by the pitch generation processing unit 49 is converted between the pitch generation processing unit 49 and the synthesis filter processing unit 132 into another pitch pattern having a different frequency (pitch modulation). Pitch modulation processing unit 13 which is then given to the synthesis filter processing unit 132
8 is provided, and in addition to the voice quality control unit 57 in FIG.
Voice quality control unit 137 for controlling not only the sampling frequency of the A / A conversion but also the modulation of the frame period and pitch at the time of synthesis.
And the voice quality control unit 137 controls the voice quality switching unit 56.
The sampling frequency in the D / A converter 53 and the synthesizing frame period in the synthesizing filter processing unit 132 are controlled, and at the same time, the pitch modulation processing unit 138 controls the pitch modulation.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 56 as in the fourth embodiment. The voice quality control unit 137 determines 11025 Hz, 8000 Hz, 1600 according to the voice quality designated by the voice quality switching unit 56.
The D / A converter 53 is controlled to perform D / A conversion at any sampling frequency of 0 Hz.

At the same time, the voice quality control section 137 determines the voice quality switching section 5
According to the voice quality designated by 6, the frame period of the synthesis performed by the synthesis filter processing unit 132 is set. The frame period of synthesis is given by the following equation.

(Frame Cycle) = (Frame Cycle during Element Creation) x (Sampling Cycle during Element Creation) / (Sampling Cycle for D / A Conversion) = (Frame Cycle During Element Creation) x (D / Sampling frequency of A conversion) / (sampling frequency at the time of segment creation) The voice quality control unit 137 further includes
(Pitch pattern given to the synthesis filter processing unit 132) = (pitch generated by the pitch generation processing unit 49) × (D / A conversion sampling period) / (at the time of segment generation) Sampling cycle) = (pitch generated by pitch generation processing unit 49) × (sampling frequency at the time of segment production) / (sampling frequency of D / A conversion) The pitch modulation processing unit 138 is controlled. .

Therefore, the same sampling frequency as the sampling frequency of the voice when the voice unit is created, that is, 110
When performing D / A conversion at 25 Hz, the voice quality control unit 137
Controls the pitch modulation processing unit 138 so that the same pitch as when the speech unit is created is given to the synthesis filter processing unit 132 as it is.

When the D / A conversion is performed at 8000 Hz, the voice quality control unit 137 multiplies the pitch at the time of creating a speech unit by (11025 [Hz] / 8000 [Hz]) and the synthesis filter processing unit 132. When performing D / A conversion at 16000 Hz (11025 [Hz] / 160
00 [Hz]), and control is given to the synthesis filter processing unit 132.

As described above, in this embodiment, the pitch given to the synthesis filter processing unit 132 is set to the voice quality control unit 137.
By performing modulation in advance in the pitch modulation processing unit 138 under the control of (1), a sampling frequency different from that at the time of creating a speech unit (second sampling period different from the first sampling period)
It is possible to cancel the change in the pitch of the synthesized voice that occurs when the D / A conversion is performed with.

Therefore, 8000 Hz or 16000 Hz, which is different from the sampling frequency at the time of creating a speech unit
Even if the D / A conversion is performed with, it is possible to obtain an analog signal of voice having the same pitch.

In any of the above-described first to ninth embodiments, the D / A converter converts the synthesized voice (discrete voice signal) output from the synthesis filter processing unit into an electrical analog signal. The case where the sampling frequency of the D / A conversion at this time is changed according to an instruction from the sound quality control unit has been described, but the present invention is not limited to this. For example, the sampling frequency itself of the synthesized voice (discrete voice signal) output from the synthesis filter processing unit may be variable.

Hereinafter, the tenth to the first, in which the sampling frequency itself of the synthesized speech (discrete speech signal) is made variable
The eighth embodiment will be described in order from the tenth embodiment.

[Tenth Embodiment] FIG. 12 is a block diagram showing the schematic arrangement of a speech analysis / synthesis apparatus according to the tenth embodiment of the present invention.

In FIG. 12, 11 is stored in the memory 141.
For the voice (discrete voice signal) sampled (sampled) at 025 Hz (first period), the frame period (first period)
Frame period), a Hanning window with a window width of 20 msec is applied at 10 msec, and the 0th to 2nd order obtained by the procedure described in the related art.
A time series (feature parameters) of parameter frames of low-order cepstrum coefficients up to the fifth order and voiced / unvoiced information of voice corresponding to each frame are stored. The low-order cepstrum coefficient has already been described in the prior art, and will be omitted here.

On the other hand, the memory 142 stores a time-series pattern of the fundamental frequency of the voice, which is obtained from the same voice by a predetermined fundamental frequency extraction method.

The synthesis filter processing unit 143 reads out each data from these two memories 141 and 142, and uses an LMA filter (not shown) in which the parameter frame (cepstral coefficient) of the voice read out from the memory 142 is used as a filter coefficient. The desired voice (discrete voice signal) is synthesized by driving with a periodic pulse based on the time-series pattern (pitch pattern) of the fundamental frequency in the voiced section and with random noise in the unvoiced section.

Since the processing up to this point is performed by the program, the voice output from the synthesis filter processing unit 143 (the LMA filter therein) is a discrete voice signal. Therefore, this discrete audio signal is supplied to the D / A converter 144,
Convert to electrical analog signal. By amplifying the analog signal of the voice thus obtained by the amplifier 145 and driving the speaker 146, a voice that can be perceptually perceived can be obtained.

The process up to this point is almost the same as the example (of the analysis / synthesis apparatus in FIG. 23) given in the prior art.

The configuration of FIG. 12 corresponds to the configuration of the analysis / synthesis apparatus of FIG. 1 according to the first embodiment. In the configuration of FIG. 1, sampling of D / A conversion by the D / A converter is performed. It is characterized in that the sampling frequency itself of the synthetic speech (discrete speech signal) is varied instead of varying the frequency.

That is, the point of the present embodiment is that the sampling frequency conversion processing for converting the sampling frequency of the synthetic speech output from the synthesis filter processing section 143 between the synthesis filter processing section 143 and the D / A converter 144. The voice quality switching unit 147 and the voice quality control unit 148 (corresponding to the voice quality switching unit 17 and the voice quality control unit 18 in FIG. 1) are provided, and the voice quality control unit 148 outputs from the synthesis filter processing unit 143. The sampling frequency conversion processing unit 149 is controlled so as to convert the sampling frequency of the synthesized voice to the frequency (second sampling period) determined by the voice quality designated by the voice quality switching unit 147.
Here, the sampling frequency (third sampling period) of the D / A conversion in the D / A converter 144 is fixed and coincides with the sampling frequency (first sampling period) of the voice when creating the cepstrum. To do.

In the present embodiment, the voice quality switching section 147.
Is the same as the voice quality switching unit 17 in FIG. 1 (specified by the user or by an application program) 3
Various types of voice quality can be designated, and the voice quality control unit 148 sets the sampling frequency of the synthesized voice output from the synthesis filter processing unit 143 to 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 147. , 12000H
The sampling frequency conversion processing unit 149 is controlled so as to convert to either Z or 10000 Hz.

Therefore, if the sampling frequency of the synthesized voice is converted to the same sampling frequency as the voice sampling frequency when the cepstrum stored in the memory 141 is created, that is, 11025 Hz, the voice is synthesized with the original voice quality. You can Converting the sampling frequency of the synthesized voice to the same sampling frequency as the voice sampling frequency at the time of creating the cepstrum is equivalent to not converting the sampling frequency, and the conversion processing in the sampling frequency conversion processing unit 149 is performed. You don't have to do it.

On the other hand, conversion to another sampling frequency (second sampling period different from the first sampling period) (11025
Converting from Hz to 12000 Hz or from 11025 Hz to 10000 Hz), as shown in FIG.
Since the effect of shifting the voice spectrum in the frequency axis direction is obtained, the individuality of the voice is changed, and the voice quality of the analog voice signal thus obtained is different from the voice quality of the original voice.

Here, the sampling frequency conversion processing unit 1
The details of the sampling frequency conversion processing by 49 will be described. Although various methods can be applied to this sampling frequency conversion, in this embodiment, it is assumed that a simple method having the configuration shown in FIG. 14 is used.

The sampling frequency conversion processing unit 149
As shown in FIG. 14A, the sampling frequency expander 149a, the low-pass filter (LPF) 149b, and the sampling frequency compressor 149c are included.

The sampling frequency expander 149a in the sampling frequency conversion processing unit 149 is supplied with the synthesized voice (voice data) output from the synthesis filter processing unit 143. It is assumed that the sampling frequency of this synthesized voice is f1.

In the sampling frequency conversion processing unit 149 of FIG. 14A, the sampling frequencies f1 to f2 = (L
/ M) f1 for frequency conversion, as shown in FIG. 14 (b), first, in the sampling frequency expander 149a,
(L-1) zero samples s0 are inserted between the samples s1 of the voice data having the sampling frequency f1.

Next, the sampling frequency expander 149a
Voice data output from, that is, between samples s1 (L
-1) voice data in which zero samples s0 are inserted,
To prevent aliasing, a low-pass filter (low-pass digital filter) 149b having a cut-off frequency of the smaller of f1 and f2 (min (f1, f2)) is used.
Pass through. Here, in order to prevent a gain decrease (1 / L times) due to the insertion of zero samples in the sampling frequency expander 149a, the low-pass filter 149b has a gain of L times in the pass band.

Finally, for the audio data that has passed through the low-pass filter 149b, the frequency compressor 149c performs a decimation process for extracting only one sample for every M samples, as shown in FIG. 14B. Audio data having a sampling frequency f2 = (L / M) f1 can be obtained.

Therefore, in the case of converting the sampling frequency from 11025 Hz to 12000 Hz as in the above example, f1 = 11025 [Hz] f2 = 12000 [Hz] f2 = (12000/11025) f1 = (160 / 147) Since it is f1, in the sampling frequency conversion processing unit 149, L = 160 M = 147 (cutoff frequency of LPF) = min (f1, f2) = f1
= 11025 [Hz], the above processing may be performed.

Similarly, the sampling frequency is set to 11025.
In the case of converting from Hz to 10000 Hz, f1 = 11025 [Hz] f2 = 10000 [Hz] f2 = (10000/1125) f1 = (400/441) f1, so the sampling frequency conversion processing unit 149 L = 400 M = 441 (cutoff frequency of LPF) = min (f1, f2) = f1
= 10000 [Hz], the above processing may be performed.

[Eleventh Embodiment] In the tenth embodiment, the voice quality switching unit 147, the voice quality control unit 148, and the sampling frequency conversion processing unit 149 are provided to easily increase the voice quality of the synthesized voice. However, the speed of the synthesized voice differs depending on the voice quality.

That is, when (sampling frequency after sampling frequency conversion processing)> (sampling frequency for D / A conversion), the speed of the synthesized voice becomes slow. On the other hand, when (sampling frequency after sampling frequency conversion processing) <(sampling frequency for D / A conversion), the speed of the synthesized voice becomes faster.

The difference in the speed of the synthesized voices is that the sampling frequency after the sampling frequency conversion processing is the same as that in the tenth embodiment (91%, 10%).
9%) does not cause much problem. However, as described below, when the sampling frequency after the sampling frequency conversion processing is significantly different from the sampling frequency at the time of D / A conversion, there is a problem.

First, if the ratio of the sampling frequency after the sampling frequency conversion processing and the sampling frequency of the D / A conversion at the time of synthesis is close to 1, the voice quality changes little, and conversely if the ratio deviates from 1, the voice quality changes greatly. To do. Therefore, if the voice quality is to be changed significantly, the ratio of these two sampling frequencies may be set to, for example, about 50% and 200%. With this, the speed of the synthesized voice is 200% and 50% of the original voice, respectively. In other words, it will be doubled and halved, making it much harder to hear.

Therefore, an eleventh embodiment will be described in which the speed of synthesized speech can be made constant even when the sampling frequency after the sampling frequency conversion processing is significantly different from the sampling frequency for D / A conversion.

FIG. 15 is a block diagram showing the schematic arrangement of the speech analysis / synthesis apparatus according to the eleventh embodiment of the present invention. The same parts as those in FIG. 12 are designated by the same reference numerals.

The configuration of FIG. 15 corresponds to the configuration of the analysis / synthesis apparatus of FIG. 3 according to the second embodiment. In the configuration of FIG. 3, sampling of D / A conversion by the D / A converter is performed. Instead of changing the frequency, the sampling frequency itself of the synthesized voice (discrete voice signal) is changed.

The point of this embodiment is that, instead of the synthesis filter processing unit 143 in FIG. 12, a synthesis filter processing unit 153 capable of controlling the frame period at the time of synthesis is provided, and the voice quality control unit 148 in FIG. Instead, a voice quality control unit 158 that controls not only the sampling frequency converted by the sampling frequency conversion processing unit 149 but also the frame period at the time of synthesis is provided, and the voice quality control unit 158 performs sampling according to the designation of the voice quality switching unit 147. Simultaneously with the sampling frequency conversion processing in the frequency conversion processing unit 149,
This is to control the frame cycle at the time of synthesis in the synthesis filter processing unit 153. Here, the sampling frequency of the D / A conversion in the D / A converter 144 (third sampling period)
Is fixed as in the tenth embodiment and matches the sampling frequency (first sampling period) of the voice when the cepstrum is created.

In this embodiment, three types of voice qualities can be designated by the voice quality switching section 147 as in the tenth embodiment. The voice quality control unit 158 determines, for example, 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 147.
The sampling frequency conversion processing unit 149 is controlled so as to convert the sampling frequency to either 8000 Hz or 16000 Hz.

At the same time, the voice quality control unit 158 controls the voice quality switching unit 1
According to the voice quality designated by 47, the frame period of the synthesis performed by the synthesis filter processing unit 153 is set.
The frame period of synthesis is given by the following equation.

(Frame period) = (Analysis frame period) × (Sampling period after sampling frequency conversion) / (D / A conversion sampling period) = (Analysis frame period) × (D / A conversion sampling frequency) / (Sampling frequency after sampling frequency conversion) Therefore, based on the above equation, when performing sampling frequency conversion processing to the same sampling frequency as D / A conversion, that is, 11025 Hz (= not performing sampling frequency conversion processing) Is the voice quality control unit 158.
Controls the synthesizing filter processing unit 153 based on the above equation so that the synthesizing frame period is 10 msec, which is the same as the frame period (first frame period) at the time of analysis (when creating the cepstrum). However, it is assumed that the cepstrum stored in the memory 141 is created under the same conditions as in the tenth embodiment.

Further, the voice quality control unit 158 converts the sampling frequency to a sampling frequency (second sampling period different from the first sampling period) different from the sampling frequency of the voice when the cepstrum is created (during analysis), for example, 8000 Hz. When processing is performed, 10 [msec] × 11025 [Hz] / 8000 [Hz] = 13.8
Control is performed so that composition is performed at a frame cycle of [msec] (second frame cycle different from the first frame cycle), and 16000H
When performing the sampling frequency conversion processing to z, 10 [msec] × 11025 [Hz] / 16000 [Hz] = 6.9
Control is performed so that composition is performed in a frame cycle of [msec] (second frame cycle different from the first frame cycle).

As described above, in this embodiment, the change in the speed of the synthesized voice when the sampling frequency conversion processing is performed to the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of analysis is synthesized. It is possible to cancel the synthesis frame period in the filter processing unit 153 by making the (second frame period) different from the frame period (first frame period) at the time of creating the cepstrum (at the time of analysis).

Therefore, even if the sampling frequency conversion processing to 8000 Hz or 16000 Hz, which is different from the sampling frequency at the time of analysis, is performed, it is possible to obtain an analog signal of voice having the same speed.

[Twelfth Embodiment] Incidentally, the first embodiment
In the embodiment of 0, when the sampling frequency conversion process is performed and then the D / A conversion is performed, the pitch of the voice, that is, the pitch of the voice is changed.

That is, when (sampling frequency after sampling frequency conversion processing)> (sampling frequency for D / A conversion), the pitch of synthesized speech becomes low. On the contrary, when (sampling frequency after sampling frequency conversion processing) <(sampling frequency for D / A conversion), the pitch of the synthesized voice becomes high.

The difference in the pitch of the synthesized voices is that the sampling frequency after the sampling frequency conversion processing is the same as the first sampling frequency as compared with the sampling frequency of the D / A conversion.
A difference of about 0 embodiment (91%, 109%) does not cause much problem.

However, if the ratio of both sampling frequencies is set to, for example, about 50% and 200% in an attempt to greatly change the voice quality, the pitch of the synthesized voice is also set to 50% and 2 respectively.
As it changes to 00%, it is the same as when creating the cepstrum 11
Compared with the voice (or the original voice) when the sampling frequency is converted to 025 Hz (or not converted), the former is synthesized by a voice whose pitch is 1 [oct] higher, and the latter is 1 [oct] lower. Since the voice is synthesized, there is a problem that it becomes difficult to hear.

Therefore, a twelfth embodiment will be described in which the pitch of synthesized speech can be made constant even when the sampling frequency after the sampling frequency conversion processing is significantly different from the sampling frequency when the cepstrum was created.

FIG. 16 is a block diagram showing the schematic arrangement of a speech analysis / synthesis apparatus according to the twelfth embodiment of the present invention. The same parts as those in FIG. 12 are designated by the same reference numerals.

The configuration of FIG. 16 corresponds to the configuration of the analysis / synthesis apparatus of FIG. 4 according to the third embodiment. In the configuration of FIG. 4, sampling of D / A conversion by the D / A converter is performed. Instead of changing the frequency, the sampling frequency itself of the synthesized voice (discrete voice signal) is changed.

The point of this embodiment is that, instead of the synthesis filter processing unit 143 in FIG. 12, a synthesis filter processing unit 163 capable of controlling the frame period at the time of synthesis is provided, and the memory 142 and the synthesis filter processing unit 163 are provided. (The time series of) the fundamental frequency read from the memory 142 during
A pitch modulation processing unit 161 that converts (pitch modulates) a pattern (pitch pattern) into another pitch pattern having a different frequency and gives it to the synthesis filter processing unit 163 is provided, and sampling is performed instead of the voice quality control unit 148 in FIG. A voice quality control unit 168 is provided for controlling not only the sampling frequency converted by the frequency conversion processing unit 149 but also the frame period and pitch modulation at the time of synthesis, and the voice quality control unit 168 is provided.
According to the designation of the voice quality switching unit 147, the sampling frequency conversion processing unit 149 controls the sampling frequency conversion processing and the synthesis filter processing unit 163 controls the frame period at the time of synthesis, and at the same time, the pitch modulation processing unit 161 controls the pitch. Is to control the modulation of. Where D / A
The sampling frequency (third sampling period) of the D / A conversion in the converter 144 is fixed as in the tenth embodiment, and coincides with the sampling frequency (first sampling period) of the voice when the cepstrum is created. It shall be.

In this embodiment, three types of voice qualities can be designated by the voice quality switching section 147 as in the tenth embodiment. The voice quality control unit 168 determines, for example, 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 147.
The sampling frequency conversion processing unit 149 is controlled so as to convert the sampling frequency to either 8000 Hz or 16000 Hz.

The voice quality control unit 168 sets the sampling frequency to be converted by the sampling frequency conversion processing unit 149 according to the voice quality specified by the voice quality switching unit 147, and at the same time, the frame period of the synthesis performed by the synthesis filter processing unit 163. To set. The frame period of synthesis is given by the following equation.

(Frame period) = (Analysis frame period) × (Sampling period after sampling frequency conversion) / (D / A conversion sampling period) = (Analysis frame period) × (D / A conversion sampling frequency) / (Sampling frequency after sampling frequency conversion) It is assumed that the cepstrum stored in the memory 141 is created under the same conditions as in the tenth embodiment.

The voice quality control section 168 further determines that the pitch (pitch pattern) given to the synthesis filter processing section 163 is (pitch given to the synthesis filter processing section 163) = (pitch read from the memory 142) × (D / A conversion) Sampling cycle) / (sampling cycle after sampling frequency conversion processing) = (pitch read from memory 142) × (sampling frequency after sampling frequency conversion processing) / (sampling frequency for D / A conversion) The pitch modulation processing unit 161 is controlled.

Therefore, when the sampling frequency conversion processing to the same sampling frequency as the D / A conversion sampling frequency, that is, 11025 Hz is performed (= no sampling frequency conversion processing is performed), the voice quality control unit 168 is used.
Controls the pitch modulation processing unit 161 based on the above equation so that the pitch given to the synthesis filter processing unit 163 becomes the same pitch as during analysis (during cepstrum creation).

Further, the voice quality control unit 168 converts the sampling frequency to a sampling frequency (second sampling period different from the first sampling period) different from the sampling frequency of the voice when the cepstrum is created (during analysis), for example, 8000 Hz. When processing is performed, the pitch read from the memory 142 is multiplied by (8000 [Hz] / 11025 [Hz]) and applied to the synthesis filter processing unit 163.
When performing the sampling frequency conversion processing to z, the pitch read from the memory 142 is set to (16000 [Hz] /
The pitch modulation processing unit 161 is controlled so as to be multiplied by 11025 [Hz]) and given to the synthesis filter processing unit 163.

As described above, in the present embodiment, the pitch given to the synthesis filter processing section 163 is set to the voice quality control section 168.
Sampling to a sampling frequency (second sampling period) different from the sampling frequency (third sampling period) at the time of D / A conversion by performing modulation in advance in the pitch modulation processing unit 161 under the control of It is possible to cancel the change in the pitch of the synthesized voice that occurs when the frequency conversion processing is performed.

Therefore, even if the sampling frequency conversion processing to 8000 Hz or 16000 Hz different from the sampling frequency at the time of D / A conversion is performed, it is possible to obtain the analog signal of the voice of the same pitch.

[Thirteenth Embodiment] FIG. 17 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the thirteenth embodiment of the present invention.

This speech rule synthesizing apparatus is realized by executing a dedicated software (sentence / speech conversion software) on an information processing apparatus such as a personal computer. The speech / speech conversion (TTS) processing function, that is, It has a sentence-speech conversion processing (sentence-speech synthesis processing) function for generating speech from a text, and the functional configuration thereof is roughly divided into a language processing unit 171 and a speech synthesis unit 172.

The language processing unit 171 analyzes the input sentence, for example, a kanji / kana mixture sentence, to generate reading information and accent information, and a phonetic symbol string in which the phoneme symbol sequence / accent information is described based on these processes. Manages the process of generating.

The voice synthesizing unit 172 controls the process of generating a voice based on the voice symbol string output from the language processing unit 171.

In the speech rule synthesizing device shown in FIG. 17, a document (Japanese document here) to be subjected to sentence-speech conversion (speech) is saved as a text file 173. According to the sentence / speech conversion software, the present apparatus reads out a sentence containing kanji and kana from the file 173 one by one, and the language processing unit 171 and the speech synthesis unit 172.
The sentence-to-speech conversion processing described below is performed to synthesize the speech.

First, the kanji / kana mixed sentence read from the text file 173 is input to the language analysis processing unit 174 in the language processing unit 171.

The language analysis processing unit 174 performs morphological analysis on the input Kanji / Kana mixed sentence and generates reading information and accent information.

For that purpose, the language analysis processing unit 174 uses the morpheme dictionary 175 having the morpheme, which is the minimum constituent element of the sentence, as an entry word, and the connection rule file 176 in which the connection rule between morphemes is registered. That is, the language analysis processing unit 1
74 finds all morpheme sequence candidates obtained by matching the input sentence with the morpheme dictionary 175 (brute force method),
Among them, the connection rule file 176 is referred to, and a combination that allows grammatical connection before and after is output. In the morpheme dictionary 175, the reading and accent types of morphemes are registered together with the grammatical information used during analysis. For this reason, if a morpheme is determined by morphological analysis, reading and accent type can be given at the same time.

For example, when a morphological analysis is performed on a sentence "go to a park and read a book", / park / f / go / book / of / read / read /.

Is divided into morphemes. Reading and accent type are given to each morpheme, and it becomes / Kouen / E / Itte / Hon / wo / Yomi / Mas /. Here, a morpheme containing "^" means that the pitch is high in the syllable immediately before and the pitch is low in the syllable immediately after. When there is no "^", it means that the accent is flat.

By the way, when a person reads a sentence, he or she does not read by adding accents in units of such morphemes, but makes a group of several morphemes and reads them by accent.

Therefore, in consideration of such a situation, the language analysis processing unit 174 further collects the morphemes with one accent phrase (unit giving an accent) and at the same time,
The movement of accents due to the compilation is also estimated. In addition to this, the language analysis processing unit 44 also adds information such as vowel devoicing and pause (breathing) during reading, and in the above example, finally generates the following reading information.

/ Cohenye / Itte. / Ho ^ o / Yomima ^ (s) / Here, the period "." Represents breath and "()" represents a syllable in which a vowel is devoiced.

Now, the language processing unit 171 is executed as described above.
When the reading information is generated by the language analysis processing unit 174 in the speech synthesis unit 172, the phoneme duration calculation processing unit 17 in the speech synthesis unit 172.
7 is activated. The phoneme duration calculation processing unit 177 determines the duration (unit: ms) of the consonant part and vowel part of each syllable included in the input sentence according to the reading information generated by the language analysis processing unit 174.

The phoneme duration calculation processing unit 177 performs the duration determination processing in such a manner that the positions of the boundaries (CV crossings) between the consonants (C) and the vowels (V) are arranged at equal intervals. It is realized by an algorithm.

The CV crossing interval is given by the speech rate control unit 178 in the voice synthesis unit 172. Although not shown, the software used in this embodiment allows the user to specify the speed of synthesized speech. Then, the speed of the voice designated by the user is given to the speech rate control unit 178, so that the speech rate control unit 178 (determined by the duration determination process in the phoneme duration calculation processing unit 177). The speed of the synthesized voice is actually changed by adjusting the interval between CVs. However, it is known from the analysis results that the duration of consonants in Japanese speech is almost constant even if the utterance speed is changed, so the duration of consonants is kept constant and the duration of vowels is adjusted. Change the interval of CV crossing.

When the phoneme duration calculation processing unit 177 determines the duration of each syllable (consonant portion and vowel portion) included in the input sentence, the pitch generation processing unit 179 in the same speech synthesis unit 172 is activated. It Pitch generation processing unit 17
Reference numeral 9 indicates the duration determined by the phoneme duration calculation processor 177, and the language analysis processor 1 (in the language processor 171).
First, the point pitch position is set based on the accent information determined by 74. Next, the pitch generation processing unit 179
Is a linear interpolation of a plurality of set point pitches.
A pitch pattern (fundamental frequency pattern) every 0 msec is obtained.

On the other hand, the phoneme parameter generation processing unit 180 in the speech synthesis unit 172 generates a phoneme parameter based on the phoneme information of the speech symbol string passed from the language analysis processing unit 174 (in the language processing unit 171). For example, the processing is performed as follows in parallel with the pitch pattern generation processing by the pitch generation processing unit 179.

First, in the present embodiment, the 0th to 25th order cepstral coefficients obtained by analyzing the real voice sampled at the sampling frequency of 11025 Hz at a window length of 20 msec and a frame period of 10 msec by the improved cepstrum method are used as consonants + vowels (CV). ), A voice unit file (not shown) in which a total of 137 voice units, which are all syllables necessary for synthesizing Japanese voice, are accumulated is prepared. The content of this speech unit file is, for example, in the main memory (not shown) at the start of the sentence-speech conversion process according to the sentence-speech conversion software.
It is assumed that the speech unit area (hereinafter, referred to as a speech unit memory) 181 that is secured in (1) is read.

The phoneme parameter generation processing unit 180, in accordance with the phoneme information in the phonetic symbol string passed from the language analysis processing unit 174 (in the language processing unit 171,) stores the above-mentioned CV-based speech units in the speech unit memory 181. Sequentially read from,
The phoneme parameters (feature parameters) of the speech to be synthesized are generated by connecting the read speech units.

When the pitch generation processing unit 179 generates a pitch pattern and the phoneme parameter generation processing unit 180 generates phoneme parameters, the synthesis filter processing unit 182 in the speech synthesis unit 172 is activated. The synthesis filter processing unit 182 is the synthesis filter processing unit 5 in FIG.
Similar to item 2, the configuration is as shown in FIG. Therefore, the synthesis filter processing unit 182 synthesizes a voice from the generated pitch pattern and phoneme parameter as follows.

First, in the voiced part (V) of the voice, the driving sound source switching part 523 switches it to the impulse generating part 522 side. The impulse generation unit 522 generates impulses at intervals according to the pitch pattern generated by the pitch generation processing unit 179, and drives the LMA filter 524 using this impulse as a sound source.

On the other hand, in the unvoiced part (U) of the voice, the driving sound source switching part 523 switches it to the white noise generating part 521 side. The white noise generator 521 generates white noise, and drives the LMA filter 524 using this white noise as a sound source. LMA filter 52
4 directly uses the cepstrum of voice as a filter coefficient.

Since the phoneme parameter generated by the phoneme parameter generation processing unit 180 in this embodiment is the cepstrum as described above, this phoneme parameter becomes the filter coefficient of the LMA filter 524 and is switched by the driving sound source switching unit 523. When driven by a sound source, it outputs synthetic speech.

The voice synthesized by the synthesis filter processing unit 172 (the LMA filter 524 therein) is a discrete voice signal, which is converted into an analog signal by the D / A converter 183 and output to the speaker 185 through the amplifier 184. For the first time, it can be heard as a sound.

The processing up to this point is almost the same as the example described in the section "Prior Art" with reference to FIG.

The configuration of FIG. 17 corresponds to the configuration of the speech rule synthesizing device of FIG. 5 according to the fourth embodiment. In the configuration of FIG. 5, D / A conversion by a D / A converter is performed. The feature is that instead of changing the sampling frequency, the sampling frequency itself of the synthesized voice (discrete voice signal) is changed.

That is, the point of the present embodiment is that the sampling frequency conversion processing for converting the sampling frequency of the synthetic speech output from the synthesis filter processing section 183 between the synthesis filter processing section 182 and the D / A converter 183. A voice quality switching unit 186 and a voice quality control unit 187 (corresponding to the voice quality switching unit 56 and the voice quality control unit 57 in FIG. 5) are provided, and the voice quality control unit 187 outputs from the synthesis filter processing unit 182. The sampling frequency conversion processing unit 188 is controlled so as to convert the sampling frequency of the synthesized voice to the frequency determined by the voice quality designated by the voice quality switching unit 186. Here, the sampling frequency (third sampling period) of the D / A conversion in the D / A converter 183 is fixed and coincides with the sampling frequency (first sampling period) of the voice when the speech unit is created. I shall.

In the present embodiment, the voice quality switching section 186.
Is the same as the voice quality switching unit 56 in FIG. 5 (specified by the user or by an application program) 3
Various types of voice quality can be designated, and the voice quality control unit 187 sets the sampling frequency of the synthetic voice output from the synthesis filter processing unit 182 to 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 186. , 12000H
The sampling frequency conversion processing unit 188 is controlled so as to convert to either Z or 10000 Hz.

Therefore, if the sampling frequency of the synthesized speech is converted to the same sampling frequency as the sampling frequency (first sampling period) when the speech unit stored in the memory 181 is created, that is, 11025 Hz (or sampling is performed). If frequency conversion processing is not performed), voice synthesis can be performed with the voice quality of the original voice.

On the other hand, if the sampling frequency conversion processing from the sampling frequency (second sampling cycle different from the first sampling cycle) different from that at the time of creating the speech unit, for example, 11025 Hz to 12000 Hz or 11025 Hz to 10000 Hz is performed, As described above, since the effect of shifting the voice spectrum in the frequency axis direction as shown in FIG. 13 is obtained, the individuality of the voice is changed, and the voice quality of the analog voice signal thus obtained is the same as that of the voice segment. Is different from the voice quality of the voice.

[Fourteenth Embodiment] In the thirteenth embodiment, the voice quality switching section 186, the voice quality control section 187, and the sampling frequency conversion processing section 188 are provided to easily increase the voice quality of synthesized speech. However, the speed of the synthesized voice differs depending on the voice quality.

That is, when (sampling frequency after sampling frequency conversion processing)> (sampling frequency for D / A conversion), the speed of synthesized speech becomes slow. On the other hand, when (sampling frequency after sampling frequency conversion processing) <(sampling frequency for D / A conversion), the speed of the synthesized voice becomes faster.

The difference in the speed of the synthesized voices is that the sampling frequency after the sampling frequency conversion processing is different from that in the thirteenth embodiment (91%, 10%).
9%) does not cause much problem. However, as described below, when the sampling frequency after the sampling frequency conversion processing is significantly different from the sampling frequency at the time of D / A conversion, there is a problem.

First, if the ratio of the sampling frequency after the sampling frequency conversion processing and the sampling frequency of the D / A conversion at the time of synthesis is close to 1, the voice quality changes little. Conversely, if the ratio deviates from 1, the voice quality changes greatly. To do. Therefore, if the voice quality is to be changed significantly, the ratio of these two sampling frequencies may be set to, for example, about 50% and 200%. With this, the speed of the synthesized voice is 200% and 50% of the original voice, respectively. In other words, it will be doubled and halved, making it much harder to hear.

Therefore, a fourteenth embodiment will be described in which the speed of synthesized speech can be made constant even when the sampling frequency after the sampling frequency conversion processing is significantly different from the sampling frequency for D / A conversion.

FIG. 18 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the fourteenth embodiment of the present invention. The same parts as those in FIG. 17 are designated by the same reference numerals.

The configuration of FIG. 18 corresponds to the configuration of the speech rule synthesizing device of FIG. 7 according to the fifth embodiment. In the configuration of FIG. 7, D / A conversion by a D / A converter is performed. Instead of varying the sampling frequency, the sampling frequency itself of the synthetic speech (discrete speech signal) is varied.

The point of this embodiment is that a speech rate control section 198 capable of controlling the speed (speech rate) of synthesized speech is provided in place of the speech rate control section 178 in FIG. 17, and the voice quality control in FIG. Instead of the section 187, a voice quality control section 197 for controlling not only the sampling frequency converted by the sampling frequency conversion processing section 188 but also the speed of synthesized speech is provided, and the voice quality control section 197 causes the voice quality switching section 186 to specify the voice quality. At the same time as the sampling frequency conversion processing in the sampling frequency conversion processing unit 188, the speed of the synthesized voice in the speech rate control unit 198 is controlled. Here, the D / A in the D / A converter 183
The sampling frequency for conversion (third sampling period) is fixed, and is the same as the sampling frequency (first sampling period) of speech at the time of creating a speech segment, as in the thirteenth embodiment.

In this embodiment, three types of voice qualities can be designated by the voice quality switching section 186, as in the thirteenth embodiment. The voice quality control unit 197 determines, for example, 11025 Hz, 8000H according to the voice quality designated by the voice quality switching unit 186.
The sampling frequency conversion processing unit 18 so as to convert the sampling frequency to either z or 16000 Hz.
8 is controlled.

At the same time, the voice quality control unit 197 determines that the voice quality switching unit 1
According to the voice quality designated by 86, the speech rate control unit 198 is controlled as follows. That is, the voice quality control unit 197
Is the above-mentioned CV crossing interval, (CV crossing interval) = (CV crossing interval when sampling frequency conversion to 11025 Hz) × (D / A conversion sampling frequency) / sampling after sampling frequency conversion The speech rate control unit 198 is controlled so that the frequency becomes).

Therefore, when converting the sampling frequency to 8000 Hz, the voice quality control section 197 converts the sampling frequency to 11025 Hz (or does not convert the sampling frequency) based on the above equation.
CV crossing interval (11025 [Hz] / 8000 [Hz])
The utterance speed control unit 198 is controlled so that the CV crossing interval is doubled. Further, the voice quality control unit 197 sets 11025H when performing sampling frequency conversion to 16000 Hz.
The utterance speed control unit 198 is controlled so that the CV crossing interval is (11025 [Hz] / 16000 [Hz]) times the CV crossing interval when the sampling frequency is converted by z (or when the sampling frequency is not converted).

As described above, in the present embodiment, by changing the interval between CVs (speech rate parameter) under the control of the speech rate control unit 198, a sampling frequency different from that at the time of segment generation (first sampling period After the sampling frequency conversion processing to a different second sampling period), the synthesis that occurs when D / A conversion is performed at the same sampling frequency (third sampling period different from the second sampling period) as when the segment was created The change in voice speed can be offset.

Therefore, even if the D / A conversion is performed at the same sampling frequency as that at the time of producing the segment after the sampling frequency conversion processing to 8000 Hz or 16000 Hz different from the sampling frequency at the time of producing the segment is performed. A voice analog signal can be obtained.

[Fifteenth Embodiment] If rule synthesis is performed based on the fourteenth embodiment, it is possible to easily realize that the speed of synthesized speech can be kept substantially constant while changing the voice quality. However, as described above, in the processing (phoneme duration calculation processing) in the phoneme duration calculation processing unit 177, the phoneme duration of consonants is made constant for each syllable even if the interval between CV crossings is changed. If the D / A conversion is performed at the same sampling frequency as that at the time of creating the segment after the synthesized speech is converted to a sampling frequency different from that at the time of creation, the duration of the consonant is shortened or extended, and as a result, the synthesized speech is This may affect clarity and naturalness.

Now, a fifteenth embodiment will be described in which the duration of the consonant of the synthesized voice can be kept constant for each syllable.

FIG. 19 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the fifteenth embodiment of the present invention. The same parts as those in FIG. 17 or 18 are designated by the same reference numerals.

The configuration of FIG. 19 corresponds to the configuration of the speech rule synthesizing device of FIG. 8 according to the sixth embodiment. In the configuration of FIG. 8, D / A conversion in the D / A converter is performed. Instead of varying the sampling frequency, the sampling frequency itself of the synthetic speech (discrete speech signal) is varied.

The point of this embodiment is that, instead of the phoneme duration calculation processing unit 177 in FIG. 18, a phoneme duration calculation processing unit 207 capable of controlling the phoneme duration is provided, and the voice quality control unit in FIG. Instead of 197, a voice quality control unit 217 that controls not only the sampling frequency converted by the sampling frequency conversion processing unit 188 but also the speed of synthesized speech is provided, and the voice quality control unit 217 causes the voice quality switching unit 186 to specify the voice quality. At the same time as the sampling frequency conversion processing in the sampling frequency conversion processing unit 188, the phoneme duration in the phoneme duration calculation processing unit 207 is controlled. Here, D / in the D / A converter 183
The sampling frequency (third sampling period) of the A conversion is fixed, and is the same as the sampling frequency (first sampling period) of the voice at the time of creating the speech unit, as in the thirteenth embodiment.

In the present embodiment, three types of voice qualities can be designated by the voice quality switching section 186 as in the fourteenth embodiment. The voice quality control unit 217 determines, for example, 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 186.
The sampling frequency conversion processing unit 188 is controlled so as to perform the sampling frequency conversion processing to the sampling frequency of either 8000 Hz or 16000 Hz.

At the same time, the voice quality control unit 217 determines the voice quality switching unit 1
In accordance with the voice quality designated by 86, the phoneme duration calculation processing unit 207 is controlled as follows. That is, the voice quality control unit 217 determines that the duration of all phonemes, that is, the duration of consonants and the duration of vowels (phoneme duration) = (phoneme duration when sampling frequency conversion to 11025 Hz) × (D / A conversion). Of the sampling frequency) / (sampling frequency after sampling frequency conversion) is controlled by the phoneme duration calculation processing unit 207.

Therefore, when performing the sampling frequency conversion processing to 8000 Hz, the voice quality control unit 217 calculates the phoneme duration (when sampling frequency conversion is performed to 11025 Hz (or when sampling frequency conversion is not performed)) based on the above equation. 11025 [Hz] / 8000 [H
z]) The phoneme duration calculation processing unit 207 is controlled so that the phoneme duration is doubled. Further, the voice quality control unit 217
When performing the sampling frequency conversion process to 00 Hz, the phoneme duration is (11025 [Hz] / 16000 [Hz]) times the phoneme duration when the sampling frequency is converted to 11025 Hz (or when the sampling frequency is not converted). Phoneme duration calculation processing unit 207
Control.

As described above, in this embodiment, the sampling frequency conversion processing is performed to the sampling frequency (the second sampling period different from the first sampling period) different from that at the time of creating the segment, and the same sampling frequency as that at the time of creating the segment ( A change in the speed of the synthesized speech that occurs when D / A conversion is performed at a third sampling period different from the second sampling period) can be offset by changing each phoneme duration at a constant rate, and The duration of the consonant of the synthesized voice can be kept constant for each syllable.

[Sixteenth Embodiment] By performing synthesis based on the fourteenth or fifteenth embodiment, the speed of synthesized speech can be kept substantially constant while changing the voice quality. However, the difference between the sampling frequency of the voice input to the D / A converter 183 (the sampling frequency after the sampling frequency conversion processing) and the sampling frequency of the D / A conversion is almost the same as the early or late rotation of the record. Therefore, it is unavoidable that the transient portion of the voice is temporally shortened or extended.

As an example, the utterance "/ wa /" is / u
The vocal tract performs a movement of opening the lips from the shape of the mouth close to / and moving to / a /. Therefore, if the sampling frequency at the time of D / A conversion is lowered and the record is delayed, this change will be gradual, and not / w /
It sounds like / uh /.

Therefore, the temporal direction of the synthesized voice transient portion generated when the D / A conversion is performed at a sampling frequency different from the sampling frequency of the voice input to the D / A converter 183 (sampling frequency after the sampling frequency conversion processing) is performed. A sixteenth embodiment in which shrinkage and extension can be suppressed will be described.

FIG. 20 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the sixteenth embodiment of the present invention. The same parts as those in FIG. 19 are designated by the same reference numerals.

The configuration of FIG. 20 corresponds to the configuration of the speech rule synthesizing device of FIG. 9 according to the seventh embodiment. In the configuration of FIG. 9, D / A conversion by a D / A converter is performed. Instead of varying the sampling frequency, the sampling frequency itself of the synthetic speech (discrete speech signal) is varied.

The point of the present embodiment is that, instead of the phoneme parameter generation processing unit 180 in FIG. 19, a phoneme parameter generation processing unit to which a function of expanding and contracting a phoneme parameter (consisting of speech units) in the time axis direction is added. A voice quality control unit 230 is provided, and instead of the voice quality control unit 217 in FIG. 19, not only the phoneme duration and the sampling frequency converted by the sampling frequency conversion processing unit 188 but also the expansion and contraction of the phoneme parameters in the time axis direction are controlled. 237 is provided, and according to the designation of the voice quality switching unit 186 by the voice quality control unit 237, the sampling frequency conversion processing and the phoneme duration calculation processing unit 20 in the sampling frequency conversion processing unit 188 is performed.
In addition to controlling the phoneme duration in 7, the phoneme parameter generation processing unit 230 is controlled, and when the transient part of the synthesized speech shrinks, the phoneme parameter is stretched in advance in the time direction to create the phoneme parameter. In the case where the transitional part of the synthetic speech is extended, the phoneme parameter is compressed in advance in the time direction to create the phoneme parameter. Here, the sampling frequency (third sampling period) of the D / A conversion in the D / A converter 183 is fixed, and the sampling frequency of the voice at the time of creating the speech unit (as in the thirteenth embodiment) ( The first sampling period).

In this embodiment, three types of voice qualities can be designated by the voice quality switching section 186 as in the fifteenth embodiment. The voice quality control unit 237 responds to the voice quality designated by the voice quality switching unit 186 by 11025 Hz (= unconverted), 800
The sampling frequency conversion processing unit 188 is controlled so as to perform the sampling frequency conversion processing to the sampling frequency of either 0 Hz or 16000 Hz.

At the same time, the voice quality control section 237 determines that the voice quality switching section 1 has
According to the voice quality designated by 86, all phoneme durations, that is, consonant durations and vowel durations, are defined as (phoneme duration) = (phoneme duration during sampling conversion to 11025Hz) × (D / The phoneme duration calculation processing unit 207 is controlled so that the sampling frequency of A conversion) / (sampling frequency after sampling frequency conversion processing).

Further, the voice quality control section 237 controls the phoneme parameter generation processing section 230, and the sampling frequency (sampling frequency after the sampling frequency conversion processing) of the synthetic speech which is an input to the D / A converter 183 and the D / A. When the transient frequency of the synthesized speech is shortened due to the different sampling frequency of the D / A conversion in the converter 183, the phonological parameter is stretched in advance in the time direction to create the phonological parameter, and When the transitional part is extended, the phoneme parameters are compressed in advance in the time direction to generate the phoneme parameters.

More precisely, the voice quality control unit 237 calculates the length of the segment itself as (Sampling frequency of D / A conversion /
Sampling frequency after sampling frequency conversion processing)
Expansion and contraction are performed so that the number is doubled, and then connection interpolation is performed to generate phoneme parameters.

That is, the voice quality control unit 23 in this embodiment.
7, the sampling frequency conversion processing unit 188 sets 1102.
Sampling frequency conversion processing to 5 Hz (sampling frequency the same as the sampling frequency when the speech unit stored in the memory 181 was created) is performed (in other words,
If the sampling frequency conversion process is not performed), the phoneme parameter is generated without expanding or contracting the speech unit, and
When performing sampling frequency conversion processing to 00,
The speech unit is extended to (11025 [Hz] / 8000 [Hz]) times and then connected and interpolated to generate phoneme parameters.
When performing the sampling frequency conversion processing to 16000 Hz, the speech unit is (11025 [Hz] / 16000 [H
z]) The phoneme parameter generation processing unit 230 is controlled so as to generate phoneme parameters by performing connection interpolation after shortening the phoneme parameter length.

As described above, in this embodiment, D / A
The sampling frequency (sampling frequency after the sampling frequency conversion process, the second sampling period) of the synthesized speech that is an input to the converter 183 and the D / A converter 183 D / A
It is possible to cancel the shrinkage or extension in the time direction of the synthetic speech transient portion caused by the different sampling frequency (third sampling period) of the A conversion by expanding or contracting the speech unit in advance when the phoneme parameter is generated. it can.

[Seventeenth Embodiment] In the fourteenth to sixteenth embodiments described above, only speech rule synthesis using a cepstrum or LPC, that is, speech rule synthesis using parameters obtained by analyzing a speech waveform is performed. Instead, it can be applied to waveform synthesis (rule synthesis by). However, in the voice rule synthesis using the parameters, the speed of the synthesized voice is kept constant while the voice quality is changed and the contraction / extension of the voice transition portion is not caused even without using the above-mentioned fourteenth to sixteenth embodiments. Various methods are applicable.

Now, a seventeenth embodiment in which this simple method is applied to voice rule synthesis using parameters will be described.

FIG. 21 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the seventeenth embodiment of the present invention. The same parts as those in FIG. 17 are designated by the same reference numerals.

The configuration of FIG. 21 corresponds to the configuration of the speech rule synthesizing apparatus of FIG. 10 according to the eighth embodiment. In the configuration of FIG. 10, the D / A conversion in the D / A converter is performed. Instead of varying the sampling frequency, the sampling frequency itself of the synthetic speech (discrete speech signal) is varied.

The point of this embodiment is that, instead of the synthesis filter processing unit 182 in FIG. 17, a synthesis filter processing unit 252 capable of controlling the frame period at the time of synthesis is provided, and the voice quality control unit 187 in FIG. Instead, a voice quality control unit 257 that controls not only the sampling frequency converted by the sampling frequency conversion processing unit 188 but also the frame period at the time of synthesis is provided, and the voice quality control unit 257 performs sampling according to the designation of the voice quality switching unit 186. Simultaneously with the sampling frequency conversion processing in the frequency conversion processing unit 188,
This is to control the frame cycle at the time of composition in the composition filter processing unit 252. Here, the sampling frequency of the D / A conversion in the D / A converter 183 (third sampling period)
Is fixed, and the sampling frequency of the voice at the time of creating the voice segment (first sampling period) as in the thirteenth embodiment.
Shall match.

In this embodiment, three kinds of voice qualities can be designated by the voice quality switching section 186, as in the thirteenth embodiment. The voice quality control unit 257 responds to the voice quality specified by the voice quality switching unit 186 by 11025 Hz (= unconverted), 800
The sampling frequency conversion processing unit 188 is controlled so as to perform the sampling frequency conversion processing to the sampling frequency of either 0 Hz or 16000 Hz.

At the same time, the voice quality control unit 257 determines that the voice quality switching unit 1
According to the voice quality designated by 86, the frame period of the synthesis performed by the synthesis filter processing unit 252 is set.
The frame period of synthesis is given by the following equation.

(Frame Cycle) = (Frame Cycle During Fragment Creation) × (Sampling Cycle After Sampling Frequency Conversion) / (D / A Conversion Sampling Cycle) = (Frame Cycle During Fragment Creation) × ( Sampling frequency of D / A conversion) / (Sampling frequency after sampling frequency conversion) Therefore, the same sampling frequency as the sampling frequency at the time of D / A conversion, that is, the sampling frequency conversion processing to 11025 Hz is performed (in other words, sampling frequency conversion (Not processed), the voice quality control unit 25
7 is a synthesis filter processing unit 25 based on the above equation so that the frame period is synthesized in the same 10 msec as when the cepstrum is created.
Control 2

Also, the voice quality control unit 257 performs a sampling frequency conversion process to a sampling frequency different from the sampling frequency of the voice at the time of creating the speech unit (second sampling period different from the first sampling period), for example, 8000 Hz. When performing, 10 [msec] × 11025 [Hz] / 8000 [Hz] = 13.8
Control is performed so that composition is performed at a frame cycle of [msec] (second frame cycle different from the first frame cycle), and 16000H
When performing the sampling frequency conversion processing to z, 10 [msec] × 11025 [Hz] / 16000 [Hz] = 6.9
Control is performed so that composition is performed in a frame cycle of [msec] (second frame cycle different from the first frame cycle).

As described above, in this embodiment, D / A
The sampling frequency (second sampling period) of the synthesized voice which is an input to the converter 183 and the D / A converter 183 D / A
The change in the synthetic speech speed that occurs when the sampling frequency for A conversion (third sampling period) is changed is calculated as the synthetic frame period (second frame period) in the synthetic filter processing unit 252 when the speech unit is created. Frame period (first
It can be offset by making it different from the frame period.

Therefore, D / A in the D / A converter 183
Even if the sampling frequency is converted to 8000 Hz or 16000 Hz different from the sampling frequency of A conversion (11025 Hz) and then D / A conversion is performed at the sampling frequency of 11025 Hz, an analog signal of voice with the same speed can be obtained. At the same time, it is possible to simultaneously prevent shrinkage or extension of the voice transition portion that occurs when D / A conversion is performed at a sampling frequency different from that at the time of creating the voice unit.

[18th Embodiment] The 13th to 1st Embodiments
Another problem exists in the seventh embodiment. When the sampling frequency (= sampling frequency after sampling frequency conversion processing) of the voice that is input to the D / A converter 183 and the sampling frequency of the D / A conversion are different, the pitch of the voice, that is, the pitch of the voice changes. It means to do. For example, when (sampling frequency after sampling frequency conversion processing)> (sampling frequency for D / A conversion), the pitch of synthesized speech becomes low. On the contrary, when (sampling frequency after sampling frequency conversion processing) <(sampling frequency for D / A conversion), the pitch of the synthesized voice becomes high.

The difference in the pitch of the synthesized voices is that the sampling frequency after the sampling frequency conversion processing is the same as that in the thirteenth embodiment (91%, 109).
%) Does not matter so much.

However, if the ratio of both sampling frequencies is set to, for example, about 50% and 200% in order to largely change the voice quality, the pitch of the synthesized voice is 200%, respectively.
It will be 50%. In this case, compared with the synthesized speech when the sampling frequency conversion processing is performed to 11025 Hz (or when the sampling frequency conversion is not performed), the former speech is synthesized with a pitch higher by 1 [oct], and the latter speech is synthesized with 1 [oct].
Since low voices are synthesized, it becomes difficult to hear.

FIG. 22 is a block diagram showing the schematic arrangement of a speech rule synthesizing apparatus according to the eighteenth embodiment of the present invention. The same parts as those in FIG. 17 are designated by the same reference numerals.

The configuration of FIG. 22 corresponds to the configuration of the speech rule synthesizing device of FIG. 11 according to the ninth embodiment. In the configuration of FIG. 11, D / A conversion by the D / A converter is performed. Instead of varying the sampling frequency, the sampling frequency itself of the synthetic speech (discrete speech signal) is varied.

The point of this embodiment is that, instead of the synthesis filter processing unit 182 in FIG. 17, a synthesis filter processing unit 272 capable of controlling the frame period at the time of synthesis and the pitch of synthesized speech is provided, and the pitch generation processing unit is provided. The pitch pattern (fundamental frequency pattern) generated by the pitch generation processing unit 179 between the 179 and the synthesis filter processing unit 272 is converted into another pitch pattern having a different frequency (pitch modulation).
The pitch modulation processing unit 278 provided to the synthesis filter processing unit 272 is provided, and in addition to the voice quality control unit 187 in FIG. 17, not only the sampling frequency converted by the sampling frequency conversion processing unit 188 but also the frame period at the time of synthesis and A voice quality control unit 277 for controlling pitch modulation is provided,
The voice quality control unit 277 causes the sampling frequency conversion processing unit 188 to perform sampling frequency conversion processing and synthesis filter processing unit 27 in accordance with the designation of the voice quality switching unit 186.
In this case, the pitch modulation in the pitch modulation processing unit 278 is controlled at the same time that the frame period of the synthesis in 2 is controlled. Here, the sampling frequency (third sampling period) of the D / A conversion in the D / A converter 183 is fixed, and the sampling frequency of the voice at the time of creating the speech unit (as in the thirteenth embodiment) ( The first sampling period).

In this embodiment, three types of voice qualities can be designated by the voice quality switching unit 186, as in the thirteenth embodiment. The voice quality control unit 277 determines, for example, 11025 Hz (= unconverted) according to the voice quality designated by the voice quality switching unit 186.
The sampling frequency conversion processing unit 188 is controlled so as to convert the sampling frequency to either 8000 Hz or 16000 Hz.

At the same time, the voice quality control unit 277 determines that the voice quality switching unit 1
The frame period of the synthesis performed by the synthesis filter processing unit 272 is set according to the voice quality designated by 86.
The frame period of synthesis is given by the following equation.

(Frame Cycle) = (Frame Cycle During Fragment Creation) × (Sampling Cycle After Sampling Frequency Conversion) / (D / A Conversion Sampling Cycle) = (Frame Cycle During Fragment Creation) × ( D / A conversion sampling frequency) / (sampling frequency after sampling frequency conversion) The voice quality control unit 277 further includes a synthesis filter processing unit 272.
Is given by: (pitch given to synthesis filter processing section 272) = (pitch when sampling frequency conversion to 11025 [Hz]) × (sampling period of D / A conversion) / (sampling frequency after sampling frequency conversion processing Period) = (pitch when sampling frequency conversion to 11025 [Hz]) × (sampling frequency after sampling frequency conversion processing) / (sampling frequency for D / A conversion) Control.

Therefore, the same sampling frequency as that at the time of D / A conversion, that is, 11025 [Hz]
When performing the sampling frequency conversion processing to (in other words, not performing the sampling frequency conversion processing),
The voice quality control unit 277 controls the pitch modulation processing unit 278 so that the pitch generated by the pitch generation processing unit 179 is given to the synthesis filter processing unit 272 as it is.

When performing the sampling frequency conversion processing to 8000 Hz, the voice quality control section 277 sets the pitch generated by the pitch generation processing section 179 to (8000 [Hz] / 1.
1025 [Hz]), and the synthesis filter processing unit 272 is controlled so that the sampling frequency conversion processing to 16000 Hz is performed, the pitch generated by the pitch generation processing unit 179 is (16000 [Hz] / 11025 [ Hz]), and control is performed so that it is given to the synthesis filter processing unit 272.

As described above, in the present embodiment, the pitch given to the synthesis filter processing section 272 is set to the voice quality control section 277.
By performing modulation in advance under the control of the D / A converter 183, the sampling frequency (second sampling period) of the synthesized voice which is an input to the D / A converter 183 and the D / A converter 183
It is possible to cancel the change in the pitch of the synthesized voice that occurs when the conversion sampling frequency (third sampling period) is changed.

Therefore, 8000 Hz or 160
After converting the sampling frequency to 00 Hz, 1102
Even if D / A conversion is performed at a sampling frequency of 5 Hz,
It is possible to obtain a voice analog signal having the same pitch.

Although the embodiments of the present invention have been described above, the present invention is not limited to these embodiments.

For example, in all the above-mentioned embodiments, the cepstrum is used as the characteristic parameter of voice,
The present invention can be applied to other parameters such as LPC and PARCOR, and similar effects can be obtained. The number of voice qualities is three in all embodiments, but may be two or four or more. Regarding the language processing unit, there is no problem even if syntax analysis or the like is inserted in addition to morphological analysis, and it is not limited to Japanese TTS and TTS of English and other languages.
Applicable to

Also, regarding the method of determining the duration, CV
The control based on a statistical method may be used instead of the method of keeping the interval constant. The present invention can also be applied to the pitch generation, not only by the point pitch method but also by using the Fujisaki model, for example.

In short, the present invention can be variously modified and implemented without departing from the scope of the invention.

[0341]

As described above in detail, according to the present invention, the sampling period of the combined discrete audio signal input to the digital / analog converting means and the discrete audio signal are converted into analog audio by the digital / analog converting means. By controlling so that the sampling period (conversion period) at the time of conversion into a signal is different, or the sampling period of the combined discrete speech signal input to the digital / analog conversion means and the discrete speech signal By performing variable control of the ratio to the sampling period (conversion period) when converting to an analog voice signal by the analog converting means according to the specified voice quality, recording of announcer utterances and recutting of voice segments are performed. It is possible to easily increase the voice quality of synthetic speech.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a voice analysis / synthesis device according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining the effect obtained by changing the sampling frequency of D / A conversion in the same embodiment.

FIG. 3 is a block diagram showing a schematic configuration of a voice analysis / synthesis device according to a second exemplary embodiment of the present invention.

FIG. 4 is a block diagram showing a schematic configuration of a voice analysis / synthesis device according to a third embodiment of the present invention.

FIG. 5 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to a fourth embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a synthesis filter processing unit.

FIG. 7 is a block diagram showing a schematic configuration of a rule synthesizing device for speech according to a fifth embodiment of the present invention.

FIG. 8 is a block diagram showing a schematic configuration of a rule synthesizing device for speech according to a sixth embodiment of the present invention.

FIG. 9 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to a seventh embodiment of the present invention.

FIG. 10 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to an eighth embodiment of the present invention.

FIG. 11 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to a ninth embodiment of the present invention.

FIG. 12 is a block diagram showing a schematic configuration of a voice analysis / synthesis device according to a tenth exemplary embodiment of the present invention.

FIG. 13 is a diagram for explaining an effect obtained by changing the sampling frequency of the synthetic voice that is an input to the D / A converter in the same embodiment.

FIG. 14 is a diagram for explaining the configuration of a sampling frequency conversion processing unit along with its operation.

FIG. 15 is a block diagram showing a schematic configuration of a voice analysis / synthesis device according to an eleventh exemplary embodiment of the present invention.

FIG. 16 is a block diagram showing a schematic configuration of a speech analysis / synthesis device according to a twelfth exemplary embodiment of the present invention.

FIG. 17 is a block diagram showing a schematic configuration of a speech rule synthesizing device according to a thirteenth embodiment of the present invention.

FIG. 18 is a block diagram showing a schematic configuration of a speech rule synthesizing device according to a fourteenth embodiment of the present invention.

FIG. 19 is a block diagram showing a schematic configuration of a speech rule synthesizing device according to a fifteenth embodiment of the present invention.

FIG. 20 is a block diagram showing a schematic configuration of a rule synthesizing device for speech according to a sixteenth embodiment of the present invention.

FIG. 21 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to a seventeenth embodiment of the present invention.

FIG. 22 is a block diagram showing a schematic configuration of a rule synthesizing device for speech according to an eighteenth embodiment of the present invention.

FIG. 23 is a block diagram showing a schematic configuration of a conventional speech analysis / synthesis device.

FIG. 24 is a block diagram showing a schematic configuration of a conventional speech rule synthesizing device.

[Explanation of symbols]

11, 141 ... Memory (feature parameter storage means), 12, 142 ... Memory (pitch pattern storage means), 13, 23, 33, 52, 112, 132, 143
153, 163, 182, 252, 272 ... Synthesis filter processing section (synthesis means), 14, 53, 144, 183 ... D / A converter (digital / analog conversion means), 17, 56, 147, 186 ... Voice quality switching Section (voice quality selection means), 18, 28, 38, 57, 67, 87, 97, 117,
137, 148, 158, 168, 187, 197, 2
17, 237, 257, 277 ... Voice quality control section, 31, 138, 161, 278 ... Pitch modulation processing section (pitch pattern modulation means), 42, 62, 72, 92, 102, 122, 172, 1
92, 202, 222, 242, 262 ... Speech synthesis section (speech synthesis means), 47, 77, 177, 207 ... Phoneme duration calculation processing section, 48, 68, 178, 198 ... Speech rate control section, 49, 179 ... Pitch generation processing unit, 50, 90, 180, 230 ... Phonological parameter generation processing unit (synthesis parameter frame time series generation means), 51, 181: Speech unit memory (unit storage unit), 149, 188 ... Sampling frequency Conversion processing unit (sample period conversion means).

Claims (132)

[Claims] 1. A storage means for storing a characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a discrete speech signal with the characteristic parameter of the speech read from the storage means as an input. And a digital / analog converting means for converting the discrete voice signal synthesized by the synthesizer into an analog voice signal at a second sampling period different from the first sampling period. Characteristic speech synthesizer. 2. A speech characteristic parameter obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in accumulating means, and the speech characteristic parameter read out from said accumulating means is input as a discrete value. A voice synthesizing method for synthesizing a voice signal, wherein the synthesized discrete voice signal is converted into an analog voice signal at a second sampling period different from the first sampling period. 3. A storage means for storing characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a discrete speech signal with the characteristic parameters of speech read out from the storage means as an input. A voice quality selection means for selectively designating the voice quality of the voice to be synthesized, and a discrete voice signal synthesized by the synthesis means determined according to the voice quality selected and designated by the voice quality selection means. A voice synthesis apparatus comprising: a digital / analog conversion means for converting into an analog voice signal at a sampling period of 2. 4. A speech characteristic parameter obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in accumulating means, and the speech characteristic parameter read out from said accumulating means is input as a discrete signal. In a voice synthesis method for synthesizing a voice signal, selection and designation of a voice quality of a voice to be synthesized is accepted, and the synthesized discrete voice signal is converted into an analog voice signal at a second sample period determined according to the accepted voice quality. A voice synthesis method characterized by: 5. From a time series of characteristic parameter frames of speech obtained by analyzing a discrete speech signal sampled at a first sampling period by applying a time window at a first frame period to the first frame period. A synthesizing means for synthesizing the discrete speech signal in a second frame cycle different from the above, and a discrete speech signal synthesized by the synthesizing means into an analog speech signal in a second sampling cycle different from the first sampling cycle. A voice synthesizer comprising: 6. From the time series of the characteristic parameter frames of the voice obtained by analyzing the discrete voice signal sampled at the first sampling period by applying a time window at the first frame period to the first frame period. A discrete voice signal is synthesized in a second frame period different from the above, and the discrete voice signal synthesized in the second frame period is converted into an analog voice signal in a second sample period different from the first sample period. A speech synthesis method characterized by the above. 7. The first frame period is calculated from a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period. , The first sampling period and the first
And a synthesizing means for synthesizing the discrete speech signal in a second frame period determined based on a second sampling period different from the sampling period of, and the discrete speech signal synthesized by the synthesizing means in the second sampling period. A voice synthesizer comprising: a digital / analog conversion means for converting into an analog voice signal. 8. The first frame period is calculated from a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period. , The first sampling period and the first
Of the second sampling period different from the sampling period of the second sampling period, the discrete speech signal is synthesized in the second frame period, and the discrete speech signal synthesized in the second frame period is analogized in the second sampling period. A speech synthesis method characterized by converting to a speech signal. 9. A storage means for storing a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period, and synthesizing. A voice quality selection means for selectively designating the voice quality of the voice to be played, and a second frame cycle determined according to the voice quality selected and designated by the voice quality selection means from the time series of the characteristic parameter frames read from the storage means. A synthesizing unit for synthesizing the discrete voice signal, and a digital / converting unit for synthesizing the discrete voice signal synthesized by the synthesizing unit into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting unit. A voice synthesizing device comprising: an analog converting means. 10. A time series of voice characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period to a storage unit. In a speech synthesis method for synthesizing a discrete speech signal from a time series of characteristic parameter frames read from the accumulating means, a selection specification of a voice quality of speech to be synthesized is accepted, and a discrete speech signal is synthesized from a time series of the characteristic parameter frame. The processing is performed in the second frame cycle determined according to the received voice quality, and the discrete voice signal synthesized in the second frame cycle is used as the analog voice in the second sample cycle determined according to the received voice quality. A speech synthesis method characterized by converting into a signal. 11. The first frame period is calculated from a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period. , Based on the first sampling period and a second sampling period different from the first sampling period, the second frame period = the first frame period × the first sampling period / the second sampling period The synthesizing means synthesizes the discrete speech signal in the second frame cycle which is defined, and the digital / analog converting means which transforms the discrete speech signal synthesized by the synthesizing means into the analog speech signal in the second sampling cycle. A speech synthesizer characterized by: 12. The first frame period is calculated from a time series of voice characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period. , Based on the first sampling period and a second sampling period different from the first sampling period, the second frame period = the first frame period × the first sampling period / the second sampling period A voice synthesizing method comprising synthesizing a discrete voice signal in a predetermined second frame period, and converting the discrete voice signal synthesized in the second frame period into an analog voice signal in the second sampling period. 13. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. Synthesizing means for synthesizing a discrete speech signal from the characteristic parameter read out from the characteristic parameter accumulating means and the fundamental frequency pattern read out from the pitch pattern accumulating means, and the discrete speech signal synthesized by the synthesizing means is converted into an analog speech signal. A digital / analog conversion means for selectively converting the analog audio signal into the first sampling period or a second sampling period different from the first sampling period. The analog conversion means and the digital / analog conversion means perform the analog sampling at the second sampling period. And a pitch pattern generating means for generating a fundamental frequency pattern different from the fundamental frequency pattern read from the pitch pattern accumulating means and applying the fundamental frequency pattern to the synthesizing means when the conversion to the voice signal is performed. A speech synthesizer. 14. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in a characteristic parameter accumulating means, and a fundamental frequency pattern of the speech is accumulated in a pitch pattern accumulating means. A voice in which a discrete voice signal is stored and synthesized from the feature parameter read from the feature parameter storage means and the fundamental frequency pattern read from the pitch pattern storage means, and the synthesized discrete voice signal is converted into an analog voice signal. In the synthesizing method, a first mode in which the conversion into the analog audio signal is performed in the first sampling period and a second mode different from the first sampling period
And a second mode performed in a sampling period of, and in the synthesis process in the second mode, a fundamental frequency pattern different from the fundamental frequency pattern read from the pitch pattern accumulating unit is generated to generate the discrete speech signal. A speech synthesis method characterized by being used for synthesizing speech. 15. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. A pitch pattern modulation unit that modulates the fundamental frequency pattern read from the pitch pattern storage unit based on the first sampling period and a second sampling period different from the first sampling period; Synthesizing means for synthesizing the discrete speech signal from the read characteristic parameter and the fundamental frequency pattern modulated by the pitch pattern modulating means, and the discrete speech signal synthesized by the synthesizing means for analog speech at the second sampling period. Speech synthesis characterized by comprising digital / analog conversion means for converting into a signal Location. 16. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is stored in a characteristic parameter storage means, and a fundamental frequency pattern of the speech is stored in a pitch pattern storage means. The basic frequency pattern that has been stored and read from the pitch pattern storage means is modulated based on the first sampling period and a second sampling period different from the first sampling period, A voice synthesizing method comprising synthesizing a discrete voice signal from the read characteristic parameter and the pitch-modulated fundamental frequency pattern, and converting the synthesized discrete voice signal into an analog voice signal at the second sampling period. 17. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. Voice quality selection means for selectively designating the voice quality of the voice to be synthesized, and pitch pattern modulation for performing modulation according to the voice quality selected and designated by the voice quality selection means for the fundamental frequency pattern read from the pitch pattern storage means. Means, a synthesizing means for synthesizing a discrete speech signal from the characteristic parameter read from the characteristic parameter accumulating means and the fundamental frequency pattern modulated by the pitch pattern modulating means, and the discrete speech signal synthesized by this synthesizing means. A second sampling period determined according to the voice quality selected and designated by the voice quality selection means. Speech synthesis apparatus characterized by comprising a digital / analog converting means for converting an analog audio signal. 18. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in a characteristic parameter accumulating means, and a fundamental frequency pattern of the speech is accumulated in a pitch pattern accumulating means. In a voice synthesizing method for synthesizing discrete voice signals based on the characteristic parameters read out from the feature parameter accumulating means and the fundamental frequency pattern read out from the pitch pattern accumulating means, selection and designation of voice quality of voice to be synthesized The target frequency pattern read from the pitch pattern accumulating means, performs modulation according to the accepted voice quality, and is discrete from the characteristic parameter read from the characteristic parameter accumulating means and the modulated fundamental frequency pattern. The voice signal is synthesized, and the synthesized discrete voice signal is Speech synthesis method characterized by converting the analog audio signal at a second sample period is determined according to only put the voice quality. 19. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. The basic frequency pattern read from the pitch pattern accumulating means is based on the first sampling period and a second sampling period different from the first sampling period (second sampling period / first sampling period). ) Double pitch pattern modulating means, synthesizing means for synthesizing a discrete voice signal from the characteristic parameter read from the characteristic parameter accumulating means and the fundamental frequency pattern modulated by the pitch pattern modulating means, and this synthesizing means. A digital / analog converter for converting the discrete voice signal synthesized by the above into an analog voice signal at the second sampling period. Speech synthesis apparatus characterized by comprising and. 20. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in a characteristic parameter accumulating means, and a fundamental frequency pattern of speech is accumulated in a pitch pattern accumulating means. The fundamental frequency pattern that has been accumulated and read from the pitch pattern accumulating means is based on the first sampling period and a second sampling period different from the first sampling period (second sampling period / second sampling period / A first sample period) times, a discrete speech signal is synthesized from the characteristic parameter read from the characteristic parameter storage means and the modulated fundamental frequency pattern, and the synthesized discrete speech signal is used as the second sample. A voice synthesizing method characterized by converting into an analog voice signal in a cycle. 21. A voice synthesizing means for synthesizing a discrete voice signal by selecting and connecting a voice segment created from a discrete voice signal sampled at a first sampling period based on given phonological information, A voice synthesizing apparatus comprising: a digital / analog converting unit for converting the discrete voice signal synthesized by the voice synthesizing unit into an analog voice signal at a second sampling period different from the first sampling period. . 22. Discrete speech signals are synthesized by selecting and connecting speech units created from discrete speech signals sampled at a first sampling period based on given phonological information, and the synthesized discrete speech signals. A voice synthesizing method comprising converting a voice signal into an analog voice signal at a second sampling period different from the first sampling period. 23. A voice synthesizing means for synthesizing a discrete voice signal by selecting and connecting a voice unit created from a discrete voice signal sampled at a first sampling period based on given phoneme information, Voice quality selection means for selectively designating the voice quality of the voice to be synthesized, and a discrete voice signal synthesized by the voice synthesis means is analogized at a second sampling period determined according to the voice quality selected and designated by the voice quality selection means. A voice synthesizing apparatus comprising: a digital / analog converting means for converting into a voice signal. 24. A voice synthesizing method for synthesizing a discrete voice signal by selecting and connecting a voice unit created from a discrete voice signal sampled at a first sampling period based on given phonological information, A voice synthesizing method, characterized in that selection and designation of a voice quality of a voice to be synthesized is accepted, and the synthesized discrete voice signal is converted into an analog voice signal at a second sampling period determined according to the accepted voice quality. 25. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is uttered in a speech to be synthesized. A voice synthesizing means for synthesizing a discrete voice signal by connecting according to a utterance speed parameter relating to a speed or a utterance time, and a digital / analog conversion for converting the discrete voice signal synthesized by the voice synthesizing means into an analog voice signal. And converting the analog audio signal into the first audio signal.
And a digital / analog converter that can selectively perform a second sample cycle different from the first sample cycle or the first sample cycle, wherein the voice synthesizer is the digital / analog converter. When the conversion to the analog voice signal is performed in the second sampling period, the speech rate parameter having a different value from that in the conversion to the analog voice signal in the first sampling period is used. A speech synthesizer characterized by being configured as follows. 26. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A voice synthesis method for synthesizing a discrete voice signal by connecting according to a voice speed parameter relating to a speed or a voice time, and converting the synthesized discrete voice signal into an analog voice signal, wherein the conversion to the analog voice signal is performed. A first mode performed in the first sampling period and a second mode different from the first sampling period
And a second mode performed at a sampling period of, and in the synthesizing process in the second mode, the utterance speed having a value different from that when the conversion into the analog audio signal is performed at the first sampling period. A speech synthesis method characterized in that a parameter is used. 27. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is uttered in a speech to be synthesized. A voice synthesizing means for synthesizing a discrete voice signal by connecting according to a speech rate parameter relating to a speed or a speech time, wherein the speech rate parameter to be used is the first sampling period and the first sampling period. And a digital / analog converting means for converting the discrete voice signal synthesized by the voice synthesizing means into an analog voice signal at the second sample period. A voice synthesizing device comprising. 28. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A speech synthesis method for synthesizing a discrete speech signal by connecting in accordance with a speech rate parameter relating to a speed or a speech time, wherein the speech rate parameter used in synthesis is the first sampling period and the first sampling period. A method for synthesizing speech, which is determined based on a second sampling period different from, and at which the synthesized discrete speech signal is converted into an analog speech signal at the second sampling period. 29. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A voice synthesizing means for synthesizing a discrete voice signal by connecting according to a utterance speed parameter related to a speed or a utterance time, a voice quality selecting means for selectively designating a voice quality of a voice to be synthesized, and a synthesizing means by the voice synthesizing means. A digital / analog converting means for converting the discrete voice signal thus converted into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means, wherein the voice synthesizing means is used. A voice synthesizer characterized in that the speech rate parameter to be determined is determined according to the voice quality selected and designated by the voice quality selecting means. Place. 30. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is uttered in a speech to be synthesized. In a speech synthesis method for synthesizing discrete speech signals by connecting according to speech rate parameters related to speed or speech time, the selection and designation of the voice quality of the speech to be synthesized is accepted, and the speech rate parameter used at the time of synthesis is A voice synthesizing method characterized by determining the received voice quality and converting the synthesized discrete voice signal into an analog voice signal at a second sampling period determined according to the received voice quality. 31. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme; and a discrete voice signal synthesized by the voice synthesizing unit into an analog voice signal. Digital / analog conversion means for converting, wherein the conversion to the analog audio signal is performed by the first
And a digital / analog conversion means capable of selectively performing a second sampling cycle different from the first sampling cycle or the first sampling cycle. When the conversion to the analog voice signal is performed at the second sampling period, the phoneme duration is set to have a phoneme duration different from that when the conversion to the analog speech signal is performed at the first sampling period. A speech synthesizer characterized by being configured to determine a time. 32. While determining the duration of each phoneme included in the given phoneme information, a speech unit created from a discrete speech signal sampled at a first sampling period is selected based on the phoneme information. A voice synthesizing method for synthesizing a discrete voice signal by connecting the selected voice units based on the determined duration of each phoneme, and converting the synthesized discrete voice signal into an analog voice signal, A first mode in which conversion into an audio signal is performed in the first sampling period and a second mode different from the first sampling period
And a second mode that is performed in the sampling cycle, and the synthesizing process in the second mode has a phoneme duration different from that when the conversion into the analog audio signal is performed in the first sampling cycle. A method for synthesizing speech, wherein the phoneme duration is determined as described above. 33. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A voice synthesizing means for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme, wherein the phoneme duration to be used is the first sampling period and A voice synthesizing unit that determines based on a second sample period different from the first sample period, and a digital that converts a discrete voice signal synthesized by the voice synthesizing unit into an analog voice signal at the second sample period. / Analog conversion means. 34. The duration of each phoneme included in the given phoneme information is determined, and a phoneme segment created from a discrete audio signal sampled at a first sampling period is selected based on the phoneme information. A speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, wherein the phoneme duration used during synthesis is the first sampling period and A voice synthesizing method characterized by determining based on a second sample period different from the first sample period, and converting the synthesized discrete voice signal into an analog voice signal at the second sample period. 35. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme; a voice quality selecting unit for selectively designating a voice quality of a voice to be synthesized; Digital / analog converting means for converting the discrete voice signal synthesized by the voice synthesizing means into an analog voice signal at a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means, The voice synthesizing means is configured to determine the phoneme duration to be used at the time of synthesis according to the voice quality selected and designated by the voice quality selecting means. A speech synthesizer. 36. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. In a speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, a selection specification of a voice quality of a speech to be synthesized is accepted and used at the time of synthesis. The phoneme duration to be determined is determined according to the received voice quality, and the synthesized discrete voice signal is converted into an analog voice signal at a second sampling period determined according to the received voice quality. Speech synthesis method. 37. The voice synthesizing means converts the analog voice signal into the analog voice signal in the first sampling cycle when the digital / analog converting means converts into the analog voice signal in the second sampling cycle. 32. The phoneme duration is determined such that the phoneme duration is (first sampling period / second sampling period) times as long as the phoneme duration when the conversion is performed. Speech synthesizer. 38. In the combining process in the second mode,
The phoneme duration is determined so that the phoneme duration is (first sample cycle / second sample cycle) times the phoneme duration when the conversion into the analog voice signal is performed in the first sample cycle. 33. The speech synthesis method according to claim 32, which is performed. 39. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is expanded or contracted in the time axis direction. A voice synthesizing unit for synthesizing a discrete voice signal by connecting, and a digital / digital converter for converting the discrete voice signal synthesized by the voice synthesizing unit into an analog voice signal at a second sampling period different from the first sampling period. A voice synthesizing device comprising: an analog converting means. 40. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is expanded or contracted in the time axis direction. A voice synthesizing method, characterized by synthesizing discrete voice signals by connecting them, and converting the synthesized discrete voice signals into analog voice signals at a second sampling period different from the first sampling period. 41. The speech synthesizing means, when connecting the selected speech unit, performs the speech unit operation for a period of time determined based on the first sampling period and the second sampling period. 40. The voice synthesizer according to claim 39, wherein the voice synthesizer is connected while expanding and contracting in the axial direction. 42. When connecting the selected speech unit, while expanding or contracting the speech unit in the time axis direction at a degree determined based on the first sampling period and the second sampling period. The voice synthesizing method according to claim 40, characterized in that connection is made. 43. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phoneme information, and the selected speech unit is expanded or contracted in the time axis direction. A voice synthesizing unit for synthesizing discrete voice signals by connecting, a voice quality selecting unit for selectively designating a voice quality of a voice to be synthesized, and a discrete voice signal synthesized by the voice synthesizing unit being selectively specified by the voice quality selecting unit. A digital / analog conversion unit for converting into an analog voice signal at a second sampling period determined according to the selected voice quality, wherein the voice synthesis unit selects and specifies the selected voice unit by the voice quality selection unit. A voice synthesizing device configured to connect while expanding and contracting in a time axis direction at a degree determined according to a voice quality. 44. A speech unit produced by selecting discrete speech signals sampled at a first sampling period based on given phonological information, and connecting the selected speech units to obtain a discrete speech signal. In the voice synthesis method for synthesizing the voice signal and converting the synthesized discrete voice signal into an analog voice signal, the selection designation of the voice quality of the voice to be synthesized is accepted, and when connecting the selected voice unit, The speech units are connected while expanding and contracting in the time axis direction at a degree determined according to the received voice quality, and when converting the synthesized discrete voice signal into an analog voice signal, the voice unit is determined according to the received voice quality. The second
A method for synthesizing speech, which comprises converting into an analog speech signal at a sampling period of. 45. The voice synthesizing means, when connecting the selected voice unit, has a size of the first voice unit (first sampling period / second sampling period) times in the time axis direction. 40. The voice synthesizing apparatus according to claim 39, wherein the voice synthesizing device is connected while being expanded and contracted. 46. When connecting the selected speech unit, the speech unit is connected while expanding and contracting in the time axis direction by a size of (first sampling period / second sampling period) times. 41. The speech synthesis method according to claim 40. 47. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames A sequence generating means, and a synthesizing means for synthesizing a discrete audio signal in a second frame cycle different from the first frame cycle from the time series of the synthetic parameter frames generated by the synthetic parameter frame time series generating means, A digitizer for converting the discrete voice signal synthesized by the synthesizer into an analog voice signal at a second sampling period different from the first sampling period. And a voice / analog conversion means. 48. A discrete voice signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of voice characteristic parameter frames obtained by analyzing the discrete voice signal by applying a time window of a first frame period. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. A discrete speech signal is synthesized in a second frame period different from the first frame period from a time series of the generated synthesis parameter frame, and the synthesized discrete speech signal is combined in a second frame period different from the first sampling period. A speech synthesis method characterized by converting into an analog speech signal at a sampling period. 49. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by analyzing the discrete speech signal by applying a time window of a first frame period. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames From the sequence generation means and the time series of the synthesis parameter frames generated by the synthesis parameter frame time series generation means, the first
A frame period, a first sampling period, and a synthesizing unit for synthesizing a discrete audio signal in a second frame period determined based on a second sampling period different from the first sampling period; A voice synthesizing apparatus, comprising: a digital / analog converting means for converting the synthesized discrete voice signal into an analog voice signal at the second sampling period. 50. A discrete voice signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of a voice characteristic parameter frame obtained by analyzing the discrete voice signal by applying a time window of a first frame period. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. Discrete in a second frame period determined based on the first frame period, the first sample period, and a second sample period different from the first sample period from the generated time series of the combined parameter frames. A voice synthesizing method comprising synthesizing a voice signal and converting the synthesized discrete voice signal into an analog voice signal at the second sampling period. 51. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of characteristic parameter frames of speech obtained by applying a time window of a first frame period to analysis. A speech unit accumulating unit for accumulating a plurality of speech units, a voice quality selecting unit for selectively designating a voice quality of a voice to be synthesized, and selecting the speech unit based on input phoneme information from the speech unit accumulating unit. A synthesized parameter frame time series generating means for connecting and generating a time series of the synthesized parameter frame, and the voice quality selecting means selectively designates from the time series of the synthesized parameter frames generated by this synthetic parameter frame time series generating means. Synthesizing means for synthesizing the discrete speech signal in the second frame period determined according to the voice quality, and the discrete speech synthesized by the synthesizing means. Speech synthesis apparatus characterized by comprising a digital / analog converting means for converting an analog audio signal at a second sample period is determined according to the voice quality selected and designated by the voice selecting unit No.. 52. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. In a voice synthesis method for synthesizing a discrete voice signal from a time series of generated synthesis parameter frames, a selection process of a voice quality of a voice to be synthesized is accepted, and a synthesis process for synthesizing a discrete voice signal from the time series of the synthesis parameter frame is performed. A second frame period determined according to the received voice quality, and the synthesized discrete voice signal is determined according to the received voice quality. A method for synthesizing speech, comprising converting into an analog speech signal at a second sampling period. 53. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by analysis by applying a time window of a first frame period. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames From the sequence generation means and the time series of the synthesis parameter frames generated by the synthesis parameter frame time series generation means, the first
Frame period, the first sampling period, and a second sampling period different from the first sampling period, the second frame period = the first frame period × the first sampling period / the second sampling period. A synthesizing means for synthesizing the discrete speech signal in the second frame period defined by the sampling period, and a digital / analog converting means for transforming the discrete speech signal synthesized by the synthesizing means into an analog speech signal in the second sampling period. A speech synthesis apparatus comprising: 54. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of a characteristic parameter frame of speech obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. From the time series of the generated combined parameter frame, the second frame period = first frame period based on the first frame period, the first sample period, and the second sample period different from the first sample period. Frame frequency × first sampling period / second sampling period, the discrete speech signal is synthesized in a second frame period, and the synthesized discrete speech signal is converted into an analog speech signal in the second sampling period. Speech synthesis method characterized by conversion. 55. Firstly, using prosody information and phonological information as input
A voice unit accumulating unit that stores a plurality of voice units created from discrete voice signals sampled at the sampling period, a pitch pattern generating unit that generates a fundamental frequency pattern of voice from the prosody information, and a phonological unit based on the phonological unit information. Phoneme parameter generating means for generating a phoneme parameter of the voice by selectively reading out and connecting the phoneme pieces from the phoneme element accumulating means, the phoneme parameter generated by the phoneme parameter generating means, and the pitch pattern generating means. A synthesizing means for synthesizing the discrete voice signal from the fundamental frequency pattern generated by the synthesizing means, and a digital / analog converting means for converting the discrete voice signal synthesized by the synthesizing means into an analog voice signal. The transformation may be the first sampling period or a first sampling period different from the first sampling period. Digital / analog conversion means capable of selectively performing two sampling periods, and the pitch pattern generation means converts the analog / speech signal into analog audio signals at the second sampling period by the digital / analog conversion means. When the conversion is performed, the basic frequency pattern is generated so that the basic frequency pattern is different from that when the conversion into the analog audio signal is performed in the first sampling period. A speech synthesizer. 56. Firstly, by inputting prosodic information and phonological information
A plurality of speech units created from discrete speech signals sampled at the sampling period of are accumulated in the speech unit accumulating unit, and a fundamental frequency pattern of speech is generated from the prosody information,
Generates phoneme parameters of the voice by selectively reading and connecting the phonemes from the phoneme storage unit based on the phoneme information, synthesizes a discrete voice signal from the generated phoneme parameters and the fundamental frequency pattern, In the voice synthesizing method of converting the synthesized discrete voice signal into an analog voice signal, a second mode different from the first mode in which the conversion into the analog voice signal is performed in the first sampling period and the first sampling period is different.
And a second mode performed at a sampling period of 1, and the fundamental frequency pattern used in the synthesis process in the second mode is different from the fundamental frequency pattern used in the synthesis process in the first mode. A speech synthesis method characterized in that a fundamental frequency pattern is generated. 57. First input with prosody information and phonological information
Based on the first sampling period and a second sampling period different from the first sampling period, the speech unit accumulating unit storing a plurality of speech units created from discrete speech signals sampled at the sampling period of And a pitch pattern generating means for generating a fundamental frequency pattern of a voice from the prosody information, and a phoneme parameter of a voice by selectively reading and connecting a voice element from the voice element accumulating means based on the phoneme information. Phonological parameter generating means, synthesizing means for synthesizing a discrete speech signal from the phonological parameter generated by the phonological parameter generating means and the fundamental frequency pattern generated by the pitch pattern generating means, and the discrete synthesizing means synthesized by the synthesizing means. Digital / analog conversion means for converting an audio signal into an analog audio signal at the second sampling period A speech synthesis apparatus comprising: 58. Firstly, by inputting prosodic information and phonological information
A plurality of speech units created from discrete speech signals sampled at the sampling period of (1) are accumulated in the speech unit accumulating unit, and the first sampling period and the second sampling period different from the first sampling period. Based on the prosody information to generate a fundamental frequency pattern of the voice, to generate a phoneme parameter of the voice by selectively reading and connecting the voice unit from the voice unit storage means based on the phoneme information, A voice synthesizing method comprising synthesizing a discrete voice signal from the generated phoneme parameter and a fundamental frequency pattern, and converting the synthesized discrete voice signal into an analog voice signal at the second sampling period. 59. Firstly, using prosody information and phonological information as input
The voice unit storing means for storing a plurality of voice units created from the discrete voice signals sampled at the sampling period, the voice quality selecting unit for selectively specifying the voice quality of the synthesized voice, and the voice quality selecting unit A pitch pattern generating means for generating a fundamental frequency pattern of voice from the prosody information according to the voice quality, and a voice by selectively reading and connecting a voice element from the voice element accumulating means based on the phonological information. And a synthesizing means for synthesizing a discrete voice signal from the phonological parameter generated by the phonological parameter generating means and the fundamental frequency pattern generated by the pitch pattern generating means. The discrete voice signal synthesized by the voice quality selecting means is adapted to the voice quality selected and designated by the voice quality selecting means. Speech synthesis apparatus characterized by comprising a digital / analog converting means for converting an analog audio signal at a second sample period defined Te. 60. Firstly, using prosody information and phonological information as input
A plurality of speech units created from discrete speech signals sampled at the sampling period of are accumulated in the speech unit accumulating unit, and a fundamental frequency pattern of speech is generated from the prosody information,
A voice that synthesizes a discrete voice signal from the generated phoneme parameter and the fundamental frequency pattern by generating a phoneme parameter of the voice by selectively reading out and connecting the phoneme unit from the voice unit storage unit based on the phoneme information. In the synthesis method, the selection and designation of the voice quality of the voice to be synthesized is accepted, the fundamental frequency pattern used in the synthesis process is generated from the prosody information according to the accepted voice quality, and the synthesized discrete voice signal is accepted. A voice synthesizing method, characterized in that the voice signal is converted into an analog voice signal at a second sampling period determined according to the voice quality. 61. The pitch pattern generation means, when the digital / analog conversion means performs conversion into an analog audio signal at the second sampling period, the analog sound signal at the first sampling period. 56. A fundamental frequency pattern having a fundamental frequency multiplied by (second sampling period / first sampling period) is generated with respect to a fundamental frequency pattern generated when the conversion to the is performed. Speech synthesizer. 62. A fundamental frequency pattern used in the synthesis process in the second mode, wherein the fundamental frequency is the first frequency
57. The voice synthesizing method according to claim 56, wherein the fundamental frequency pattern used for the synthesizing process in the mode is generated so as to be (second sampling period / first sampling period) times the fundamental frequency. 63. Accumulating means for accumulating voice characteristic parameters obtained by analyzing a discrete voice signal sampled at a first sampling period, and a discrete voice signal using the voice characteristic parameters read from the storage means as an input. And a sampling period converting unit for converting the sampling period of the discrete speech signal synthesized by the combining unit into a second sampling period different from the first sampling period, and the sampling period converting unit. A voice synthesizing apparatus comprising: a digital / analog conversion means for converting a discrete voice signal whose sample period has been converted into an analog voice signal at a third sample period different from the second sample period. 64. A voice characteristic parameter obtained by analyzing a discrete voice signal sampled at a first sampling period is stored in a storage means, and the voice characteristic parameter read from the storage means is input as a discrete signal. In a speech synthesis method for synthesizing a speech signal, a sampling period of the synthesized discrete speech signal is converted into a second sampling period different from the first sampling period, and after the sampling period conversion into the second sampling period. The method of synthesizing speech, wherein the discrete speech signal of is converted into an analog speech signal at a third sampling period different from the second sampling period. 65. Accumulating means for accumulating voice characteristic parameters obtained by analyzing a discrete speech signal sampled at a first sampling period, and a discrete speech signal using the speech characteristic parameters read from the accumulating means as an input. Synthesizing means, a voice quality selecting means for selectively designating the voice quality of the voice to be synthesized, and a sampling period of the discrete speech signal synthesized by the synthesizing means according to the voice quality selected and designated by the voice quality selecting means. Sampling period conversion means for converting to a predetermined second sampling period, and the discrete speech signal whose sampling period has been converted by this sampling period conversion device are different from the second sampling period (however, the second sampling period is (Except when it matches the first sampling period)
A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 66. A voice characteristic parameter obtained by analyzing a discrete voice signal sampled at a first sampling period is stored in a storage means, and the voice feature parameter read from the storage means is input as a discrete signal. In a voice synthesizing method for synthesizing a voice signal, a selection and designation of a voice quality of a voice to be synthesized is accepted, and a sample period of the synthesized discrete voice signal is converted into a second sample period determined according to the accepted voice quality. , The discrete speech signal after the conversion of the sampling period into the second sampling period is different from the second sampling period (except when the second sampling period matches the first sampling period) A method for synthesizing speech, which comprises converting into an analog speech signal at a sampling period of. 67. From a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period to the first frame period. A synthesizing means for synthesizing the discrete speech signal in a second frame period different from the above, and a sampling period of the discrete speech signal synthesized by the synthesizing means into a second sampling period different from the first sampling period. Sample period conversion means and digital / analog conversion means for converting the discrete audio signal whose sample cycle is converted by the sample cycle conversion means into an analog audio signal at a third sample cycle different from the second sample cycle. A voice synthesizing device comprising. 68. From a time series of characteristic parameter frames of speech obtained by analyzing a discrete speech signal sampled at a first sampling period by applying a time window at a first frame period, the first frame period A discrete voice signal is synthesized in a second frame period different from, and a sample period of the discrete voice signal synthesized in the second frame period is converted into a second sample period different from the first sample period, A voice synthesizing method characterized in that the discrete voice signal after the conversion of the sample period into the second sample period is converted into an analog voice signal at a third sample period different from the second sample period. 69. From a time series of speech characteristic parameter frames obtained by analyzing a discrete speech signal sampled at a first sampling period by applying a time window at a first frame period to the first frame period. , A second sampling period different from the first sampling period and a third sampling period different from the second sampling period
And a synthesizing unit for synthesizing the discrete speech signal in the second frame period determined based on the sampling period of, and a sampling period conversion for transforming the sampling period of the discrete speech signal synthesized by the synthesizing unit into the second sampling period. A voice synthesizing apparatus comprising: means and a digital / analog conversion means for converting the discrete voice signal whose sample period is converted by the sample period converter into an analog voice signal at the third sample period. 70. From the time series of characteristic parameter frames of speech obtained by analyzing a discrete speech signal sampled at a first sampling period by applying a time window at a first frame period to the first frame period. A second sampling period different from the first sampling period and a third sampling period different from the second sampling period
A discrete speech signal is synthesized at a second frame period determined based on the sampling period of, and the sampling period of the discrete speech signal synthesized at the second frame period is converted to the second sampling period. The voice synthesis method, comprising: converting the discrete voice signal after the conversion of the sample period to the sample period into the analog voice signal at the third sample period. 71. Accumulating means for accumulating a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period, and combining. Voice quality selection means for selectively designating the voice quality of the voice to be played, and a second frame cycle determined according to the voice quality selected and designated by the voice quality selection means from the time series of the characteristic parameter frames read from the storage means. A synthesizing unit for synthesizing the discrete speech signal, and a sampling period conversion for transforming the sampling period of the discrete speech signal synthesized by the synthesizing unit into a second sampling period determined according to the voice quality selected and designated by the voice quality selecting unit. Means and the discrete speech signal whose sampling period is converted by the sampling period converting means is different from the second sampling period (where the second sampling period is the first sampling period). Except when it matches the sampling period of
A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 72. A time series of voice characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period to a storage unit. In a speech synthesis method for synthesizing a discrete speech signal from a time series of characteristic parameter frames read from the storage means, a selection specification of a voice quality of speech to be synthesized is accepted, and a discrete speech signal is synthesized from a time series of the characteristic parameter frame. The processing is performed in the second frame cycle determined according to the received voice quality, and the sampling cycle of the discrete voice signal synthesized in the second frame cycle is determined in the second sample cycle determined according to the received voice quality. And the discrete speech signal after the conversion of the sampling period into the second sampling period is different from the second sampling period (where the second sampling period is the first sampling period). Speech synthesis method characterized by converting the excluding) analog audio signal at a third sample periods if it matches the sample period. 73. From a time series of speech characteristic parameter frames obtained by analyzing a discrete voice signal sampled at a first sampling period by applying a time window at a first frame period, the first frame period A second sampling period different from the first sampling period and a third sampling period different from the second sampling period
Second frame period = first frame period × second sample period / third sample period based on the sample period of Sampling period converting means for converting the sampling period of the discrete speech signal synthesized by the synthesizing means into the second sampling period, and the discrete speech signal having the sampling period transformed by the sampling period converting means for the third sampling period. And a digital / analog conversion means for converting into an analog voice signal according to 1. 74. From a time series of speech characteristic parameter frames obtained by analyzing a discrete speech signal sampled in a first sampling period by applying a time window in a first frame period, the first frame period , A second sampling period different from the first sampling period and a third sampling period different from the second sampling period
2nd frame period = first frame period × second sample period / third sample period based on the sample period of The sampling period of the discrete speech signal synthesized in the frame period is converted into the second sampling period, and the discrete speech signal after the sampling period conversion into the second sampling period is converted into the analog speech signal in the third sampling period. A speech synthesis method characterized by conversion. 75. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. Synthesizing means for synthesizing a discrete speech signal from the characteristic parameter read out from the characteristic parameter accumulating means and the fundamental frequency pattern read out from the pitch pattern accumulating means, and the sampling period of the discrete speech signal synthesized by the synthesizing means is converted. And a second sampling period conversion means for converting the first sampling period or a second sampling period different from the first sampling period.
To a sampling period, and a discrete speech signal whose sampling period has been converted by the sampling period converting unit is a third sample different from the second sampling period. A digital / analog conversion means for converting into an analog voice signal at a cycle, and a fundamental frequency read from the pitch pattern accumulating means when the sampling cycle conversion means performs the sampling cycle conversion to the second sampling cycle. A voice synthesizing apparatus comprising: a pitch pattern generating means for generating a fundamental frequency pattern different from a pattern and giving it to the synthesizing means. 76. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in a characteristic parameter accumulating means, and a fundamental frequency pattern of speech is accumulated in a pitch pattern accumulating means. A voice synthesizing method for synthesizing discrete voice signals from the characteristic parameters read from the feature parameter storage means and the fundamental frequency pattern read from the pitch pattern storage means, wherein the sampling period of the synthesized discrete voice signals is A first mode in which the first sampling period is maintained and a second mode in which the second sampling period is different from the first sampling period are prepared, and the pitch pattern is set in the second mode. A fundamental frequency pattern different from the fundamental frequency pattern read from the storage means is generated and used for the synthesis of the discrete speech signal. , The synthesized sampling period of the discrete speech signal is converted into the second sampling period, and the discrete speech signal after the sampling period conversion into the second sampling period is converted into a third sampling period different from the second sampling period. A voice synthesis method characterized by converting to an analog voice signal at a sampling period. 77. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. Pitch pattern modulating means for modulating the fundamental frequency pattern read from the pitch pattern accumulating means based on a second sampling cycle different from the first sampling cycle and a third sampling cycle different from the second sampling cycle. A synthesizing means for synthesizing a discrete speech signal from the characteristic parameter read from the characteristic parameter accumulating means and the fundamental frequency pattern modulated by the pitch pattern modulating means, and a sample of the discrete speech signal synthesized by this synthesizing means. A sampling cycle converting means for converting the cycle into the second sampling cycle, and the sampling cycle converting means Speech synthesis apparatus characterized by the period; and a digital / analog converting means for converting the discrete audio signal converted into the third analog audio signal at a sampling period of the. 78. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is accumulated in a characteristic parameter accumulating means, and a fundamental frequency pattern of speech is accumulated in a pitch pattern accumulating means. The basic frequency pattern read out from the pitch pattern storage means is modulated based on a second sampling period different from the first sampling period and a third sampling period different from the second sampling period. , A discrete speech signal is synthesized from the characteristic parameter read from the characteristic parameter accumulating means and the pitch-modulated fundamental frequency pattern, and a sampling period of the synthesized discrete speech signal is converted into the second sampling period. A sound characterized by converting a discrete speech signal after conversion of a sampling period into two sampling periods into an analog speech signal at the third sampling period. Synthetic methods. 79. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. Voice quality selection means for selectively designating the voice quality of the voice to be synthesized, and pitch pattern modulation for performing modulation according to the voice quality selected and designated by the voice quality selection means for the fundamental frequency pattern read from the pitch pattern storage means. Means, a synthesizing means for synthesizing a discrete speech signal from the characteristic parameter read from the characteristic parameter accumulating means and the fundamental frequency pattern modulated by the pitch pattern modulating means, and a synthesizing means for synthesizing the discrete speech signal by the synthesizing means. A second sampling period is determined according to the voice quality selected and designated by the voice quality selection means. Sampling period conversion means for converting into a sampling period, and the discrete speech signal whose sampling period has been converted by this sampling period converting means is different from the second sampling period (however, the second sampling period is the first sampling period). Except when it matches
A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 80. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is stored in a characteristic parameter storage means, and a fundamental frequency pattern of the speech is stored in a pitch pattern storage means. A voice synthesis method for synthesizing discrete voice signals based on the characteristic parameters read out from the feature parameter storage means and the fundamental frequency pattern read out from the pitch pattern storage means. The target frequency pattern read from the pitch pattern accumulating means, performs modulation according to the accepted voice quality, and discrete from the characteristic parameter read from the characteristic parameter accumulating means and the modulated fundamental frequency pattern. A sample of the synthesized discrete speech signal Period is converted into a second sampling period which is determined according to the received voice quality, and the discrete speech signal after the sampling period conversion to this second sampling period is different from the second sampling period (however, (Except when the second sampling period matches the first sampling period) A voice synthesis method characterized by converting to an analog voice signal at the third sampling period. 81. A characteristic parameter accumulating means for accumulating characteristic parameters of speech obtained by analyzing a discrete speech signal sampled at a first sampling period, and a pitch pattern accumulating means for accumulating a fundamental frequency pattern of speech. The fundamental frequency pattern read from the pitch pattern accumulating unit is based on a second sampling period different from the first sampling period and a third sampling period different from the second sampling period (third sampling period). (Sampling period / second sampling period) times, a pitch pattern modulating unit, a characteristic parameter read from the characteristic parameter accumulating unit, and a fundamental frequency pattern modulated by the pitch pattern modulating unit to synthesize a discrete audio signal. Synthesizing means for converting the sampling period of the discrete speech signal synthesized by the synthesizing means into the second sampling period. Means a speech synthesis device is sampled periodically by the sample periods converter characterized by comprising a digital / analog converting means for converting the analog audio signal a discrete audio signal converted by the third sample periods. 82. A characteristic parameter of speech obtained by analyzing a discrete speech signal sampled at a first sampling period is stored in a characteristic parameter storage means, and a fundamental frequency pattern of the speech is stored in a pitch pattern storage means. The basic frequency pattern that has been accumulated and read out from the pitch pattern accumulating unit has a second sampling period different from the first sampling period and a third sampling period different from the second sampling period. (Third sampling period / second sampling period) times, a discrete audio signal is synthesized from the characteristic parameter read from the characteristic parameter accumulating means and the modulated fundamental frequency pattern, and the synthesized discrete The sampling period of the audio signal is converted into the second sampling period, and the discrete audio signal after the conversion of the sampling period into the second sampling period is analyzed in the third sampling period. Speech synthesis method characterized by converting the audio signal. 83. A speech segment created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information and connected to synthesize a discrete speech signal, and the synthesized speech signal is synthesized. A voice synthesizing unit for converting a sample period of the voice signal into a second sample period different from the first sample period, and a discrete voice signal synthesized by the voice synthesizing unit and having the sample period converted into the second sample period. A voice synthesizing apparatus comprising: a digital / analog converting means for converting into an analog voice signal at a third sampling period different from the sampling period. 84. A voice unit produced from a discrete voice signal sampled at a first sampling period is selected and connected based on given phoneme information to synthesize a discrete voice signal, and the synthesized discrete voice signal. The sampling period of the audio signal is converted into a second sampling period different from the first sampling period, and the discrete audio signal after the sampling period conversion into the second sampling period is different from the second sampling period. A voice synthesizing method characterized by converting into an analog voice signal in a third sampling period. 85. A voice quality selecting means for selectively specifying a voice quality of a voice to be synthesized, and a voice unit created from a discrete voice signal sampled at a first sampling period are selected based on given phoneme information. A voice synthesis means for synthesizing the discrete voice signals by connecting them to each other, wherein the sample period of the synthesized discrete voice signals is converted into a second sample period determined according to the voice quality selected and designated by the voice quality selection means. And the discrete speech signal output from the speech synthesis means is different from the second sampling period (where the second sampling period is the first sampling period).
(Excluding the case where the sampling period is equal to the sampling period), the speech synthesizing device comprising: a digital / analog converting means for converting into an analog speech signal at a third sampling period. 86. A voice synthesizing method for synthesizing a discrete voice signal by selecting and connecting a voice unit created from a discrete voice signal sampled at a first sampling period based on given phoneme information, The selection and designation of the voice quality of the voice to be synthesized is accepted, the sampling period of the synthesized discrete voice signal is converted into the second sampling period determined according to the accepted voice quality, and the sampling to the second sampling period is performed. The discrete voice signal after the period conversion is converted into an analog voice signal at a third sampling period which is different from the second sampling period (except when the second sampling period matches the first sampling period). A speech synthesis method characterized by the above. 87. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A voice synthesizing means for synthesizing a discrete voice signal by connecting in accordance with a voice speed parameter relating to a speed or a voice time, wherein a sampling period of the synthesized discrete voice signal is different from the first sampling period. A voice synthesizing unit capable of being converted into two sample periods, and a digital unit for converting the discrete voice signal output from the voice synthesizing unit into an analog voice signal at a third sample period different from the second sample period. / Analog conversion means, the speech synthesis means, when performing the synthesis process when performing the sampling period conversion to the second sampling period, a value different from that when the conversion of the sampling period is not performed. Speech synthesis apparatus characterized by being configured to use the serial speech rate parameter. 88. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A speech synthesis method for synthesizing a discrete speech signal by connecting according to a speech rate parameter related to a speech rate or a speech time, wherein a sampling period of the synthesized discrete speech signal is kept at the first sampling period. And a second mode in which a second sampling period different from the first sampling period is prepared, and in the second mode, the speech rate parameter having a value different from that in the first mode. Is used to synthesize the discrete speech signal, the sampling period of the synthesized discrete speech signal is converted into the second sampling period, and the discrete speech signal after the sampling period conversion to the second sampling period is converted into the second sampling period. First Speech synthesis method characterized in that of the sampling period into an analog audio signal at a different third sample periods. 89. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is spoken of a speech to be synthesized. A voice synthesizing means for synthesizing a discrete voice signal by connecting according to a voice speed parameter relating to a speed or a voice time, wherein a sampling period of the synthesized discrete voice signal is different from the first sampling period. A voice synthesizing unit capable of being converted into two sample periods, and a digital unit for converting the discrete voice signal output from the voice synthesizing unit into an analog voice signal at a third sample period different from the second sample period. / Analog conversion means, and the voice synthesis means is configured to determine the speech rate parameter to be used based on the second sampling period and the third sampling period. Speech synthesis apparatus characterized by being. 90. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is uttered in a speech to be synthesized. A speech synthesis method for synthesizing a discrete speech signal by connecting according to a speech rate parameter related to a speech rate or a speech time, wherein a sampling period of the synthesized discrete speech signal remains the first sampling period. And a second mode that is a second sampling period different from the first sampling period, and in the second mode, the speech rate parameter to be used is set to the second sampling period and the second sampling period. After deciding on the basis of a third sampling period different from the second sampling period to synthesize the discrete speech signal, the sampling period of the synthesized discrete speech signal is converted to the second sampling period, Speech synthesis method characterized by converting the analog audio signal a discrete audio signal after the sampling cycle conversion into 2 sample periods in the third sample periods. 91. A voice quality selection means for selectively designating a voice quality of a voice to be synthesized, and a voice unit created from a discrete voice signal sampled at a first sampling period are selected based on given phoneme information. Then, a voice synthesizing means for synthesizing a discrete voice signal by connecting the selected voice unit according to a utterance speed parameter related to a utterance speed or a utterance time of a voice to be synthesized, wherein the synthesized discrete A voice synthesizing unit for converting a sample period of the voice signal into a second sample period determined according to the voice quality selected and designated by the voice quality selecting unit, and a discrete voice signal output from the voice synthesizing unit for the second voice Different from the sampling period (however, the second sampling period is the first
A digital / analog conversion unit for converting into an analog voice signal at a third sample period), the voice synthesizing unit determines the speech rate parameter to be used by the voice quality selecting unit. A voice synthesizing device, characterized in that it is configured to decide according to a voice quality selected and designated. 92. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phoneme information, and the selected speech unit is spoken of a speech to be synthesized. In a speech synthesis method for synthesizing a discrete speech signal by connecting according to a speech rate parameter related to a speech rate or a speech time, a selection and designation of a voice quality of speech to be synthesized is accepted, and the speech rate parameter to be used is accepted. And synthesizes the discrete voice signal by converting the sample period of the synthesized discrete voice signal into a second sample period determined according to the received voice quality. A third sample period, which is different from the second sample period (except when the second sample period matches the first sample period); A voice synthesis method characterized by converting into an analog voice signal during a period. 93. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme, wherein a sampling period of the synthesized discrete voice signal is the first A voice synthesizer capable of converting into a second sample period different from the sample period, and a discrete voice signal output from the voice synthesizer with a third sample period different from the second sample period. A digital / analog converting means for converting into a voice signal, wherein the voice synthesizing means performs sampling cycle conversion to the second sampling cycle, Speech synthesis apparatus characterized by being configured to make a determination of the phoneme duration so as to be different phoneme duration. 94. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, wherein the sampling period of the synthesized discrete speech signal is the first sampling period. A first mode to be left as it is and a second mode to be a second sampling period different from the first sampling period are prepared, and when the second mode is different from the first mode. The phoneme duration is determined so that it becomes a phoneme duration and is used in the synthesis of the discrete speech signal, the sampling period of the synthesized discrete speech signal is converted into the second sampling period, and the second sampling period Mark to Speech synthesis method characterized by converting the analog audio signal at a different third sample periods and a discrete audio signal after cycle conversion the second sample period. 95. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme, wherein the sampling period of the synthesized discrete voice signal is the first A voice synthesizer capable of converting into a second sample period different from the sample period, and a discrete voice signal outputted from the voice synthesizer with a third sample period different from the second sample period. A digital / analog converting means for converting into a voice signal, wherein the voice synthesizing means determines the phoneme duration to be used at the time of synthesizing based on the second sampling period and the third sampling period. Speech synthesis apparatus characterized by being configured to. 96. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. In the speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, the phonological duration used during synthesis is defined as the first sampling period. Is determined based on a different second sampling period and a third sampling period different from the second sampling period, and the sampling period of the synthesized discrete audio signal is converted into the second sampling period, A voice synthesizing method, characterized in that a discrete speech signal after conversion of a sampling period into a second sampling period is converted into an analog speech signal at the third sampling period. 97. A voice quality selection means for selectively designating a voice quality of a voice to be synthesized, a duration of each phoneme included in given phoneme information is determined, and a discrete voice sampled at a first sampling period. Select a speech unit created from a signal based on the phonological information,
A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the duration of each of the determined phonemes, the sampling period of the synthesized discrete voice signal being determined by the voice quality selecting unit. A voice synthesizing means for converting into a second sampling period determined according to the selected and designated voice quality, and a discrete voice signal outputted from this voice synthesizing means are different from the second sampling period (however, the second sampling period is different). First sampling period
(Excluding the case where it matches the sampling period of 1), the digital / analog converting means for converting into an analog voice signal at a third sampling period, wherein the voice synthesizing means selects the phoneme duration to be used at the time of synthesizing the voice quality. A voice synthesizing apparatus, characterized in that it is configured to decide according to the voice quality selected and designated by the means. 98. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. In a speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, the selection of the voice quality of the speech to be synthesized is accepted and used. After the phoneme duration is determined according to the accepted voice quality and the discrete voice signal is synthesized, the sampling period of the synthesized discrete voice signal is converted into a second sampling period determined according to the accepted voice quality. , The discrete speech signal after the conversion of the sampling period into the second sampling period is different from the second sampling period (except when the second sampling period matches the first sampling period) A method for synthesizing speech, which comprises converting into an analog speech signal at a sampling period of. 99. The duration of each phoneme included in the given phoneme information is determined, while a phoneme segment created from a discrete audio signal sampled at a first sampling period is selected based on the phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected speech units based on the determined duration of each phoneme, wherein a sampling period of the synthesized discrete voice signal is the first A voice synthesizer capable of converting into a second sample period different from the sample period, and a discrete voice signal output from the voice synthesizer with a third sample period different from the second sample period. A digital / analog converting means for converting into a voice signal, wherein the voice synthesizing means performs a synthesizing process at the time of performing the sample period conversion to the second sample period when the sample period is not converted. It is determined in the phoneme duration (second sample period /
A speech synthesizer characterized by being configured to use a phoneme duration that is three times the third sampling period). 100. The duration of each phoneme included in the given phoneme information is determined, while the phoneme segment created from the discrete audio signal sampled at the first sampling period is selected based on the phoneme information. A speech synthesis method for synthesizing a discrete speech signal by connecting the selected speech units based on the determined duration of each phoneme, wherein the sampling period of the synthesized discrete speech signal is the first sampling period. A first mode to be left as it is and a second mode to be a second sampling period different from the first sampling period are prepared, and in the second mode, the second sampling period and the second sampling period are set. A third sampling period different from the second sampling period is used, and the phoneme duration (second sampling period / third sampling period) determined in the first mode is used.
The phoneme duration is determined so as to be doubled and used for the synthesis of the discrete speech signal, the sampling period of the synthesized discrete speech signal is converted into the second sampling period, and the sampling period of the second sampling period is changed to the second sampling period. A voice synthesizing method comprising converting a discrete voice signal after sample period conversion into an analog voice signal at the third sample period. 101. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is expanded or contracted in the time axis direction. A voice synthesizing means for synthesizing a discrete voice signal by connecting the voice synthesizer capable of converting a sample period of the synthesized discrete voice signal into a second sample period different from the first sample period. Means for converting the discrete voice signal output from the voice synthesizer into an analog voice signal at a third sampling period different from the second sampling period. Speech synthesizer. 102. A speech unit created from a discrete speech signal sampled at a first sampling period is selected based on given phonological information, and the selected speech unit is expanded or contracted in the time axis direction. The discrete speech signal is synthesized by connecting, the sampling period of the synthesized discrete speech signal is converted into a second sampling period different from the first sampling period, and the sampling period is converted into the second sampling period. A voice synthesizing method, characterized in that the subsequent discrete voice signal is converted into an analog voice signal at a third sampling period different from the second sampling period. 103. The voice synthesizing means, when connecting the selected speech unit, performs the speech unit operation for a period of time determined based on the second sampling period and the third sampling period. The voice synthesis device according to claim 101, wherein the voice synthesis device is connected while expanding and contracting in the axial direction. 104. When connecting the selected speech unit, the speech unit is connected to the second sampling period and the third sampling period.
102. The connection is made while expanding and contracting in the time axis direction at a degree determined based on the sampling cycle of.
The described speech synthesis method. 105. A voice quality selection means for selectively designating a voice quality of a voice to be synthesized, and a voice unit created from a discrete voice signal sampled at a first sampling period based on given phoneme information. A voice synthesizing unit for synthesizing a discrete voice signal by connecting the selected voice unit while expanding and contracting in the time axis direction at a degree determined according to the voice quality selected and designated by the voice quality selecting unit, A voice synthesizing unit for converting the sample period of the synthesized discrete voice signal into a second sample period determined according to the voice quality selected and designated by the voice quality selecting unit, and a discrete voice signal output from the voice synthesizing unit. Different from the second sampling period (where the second sampling period is the first
(Excluding the case where the sampling period is equal to the sampling period), the speech synthesizing device comprising: a digital / analog converting means for converting into an analog speech signal at a third sampling period. 106. A discrete speech signal is selected by selecting a speech unit created from a discrete speech signal sampled at a first sampling period based on given phonological information and connecting the selected speech units. In the voice synthesis method for synthesizing the voice signal and converting the synthesized discrete voice signal into an analog voice signal, the selection designation of the voice quality of the voice to be synthesized is accepted, and when connecting the selected voice unit, The speech units are connected while expanding and contracting in the time axis direction at a degree determined according to the received voice quality, and when the discrete voice signal is synthesized, the sampling period of the discrete voice signal is changed according to the received voice quality. Is converted to a second sampling period determined by the above, and the discrete speech signal after the conversion of the sampling period into the second sampling period is different from the second sampling period (where the second sampling period is the first sampling period). Speech synthesis method characterized by converting the excluding) analog audio signal at a third sample periods if it matches the sample period. 107. The speech synthesis means, when connecting the selected speech unit, has a size (second sampling period / third sampling period) times as large as the speech unit in the time axis direction. The voice synthesizing device according to claim 103, wherein the voice synthesizing device is connected while being expanded and contracted. 108. When the selected speech unit is connected, the speech unit is connected in the time axis direction (second sampling period /
105. The speech synthesis method according to claim 104, wherein connection is performed while expanding and contracting by a size equal to a third sampling period). 109. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by analysis by applying a time window of a first frame period. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames A sequence generating means, and a synthesizing means for synthesizing a discrete audio signal in a second frame cycle different from the first frame cycle from the time series of the synthetic parameter frames generated by the synthetic parameter frame time series generating means, A sampling period converter for converting the sampling period of the discrete speech signal synthesized by the synthesizing unit into a second sampling period different from the first sampling period. And a digital / analog converting means for converting the discrete voice signal whose sample period is converted by the sample period converting means into an analog voice signal at a third sample period different from the second sample period. A speech synthesizer characterized by. 110. A discrete speech signal sampled at a first sampling period is clipped in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. A discrete speech signal is synthesized in a second frame period different from the first frame period from the time series of the generated synthesis parameter frames, and the sampling period of the discrete speech signal synthesized in the second frame period is the first frame period. To a second sample period different from the second sample period, and the discrete speech signal after the sample period conversion to the second sample period is analyzed by a third sample period different from the second sample period. Speech synthesis method characterized by converting the audio signal. 111. A discrete voice signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of a voice characteristic parameter frame obtained by analyzing the discrete voice signal by applying a time window of a first frame period. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames From the sequence generation means and the time series of the synthesis parameter frames generated by the synthesis parameter frame time series generation means, the first
Of the first sampling period, a second sampling period different from the first sampling period, and a third frame period different from the second sampling period to synthesize a discrete audio signal in a second frame period. Synthesizing means, sampling period converting means for converting the sampling period of the discrete speech signal synthesized by the synthesizing means into the second sampling period, and the discrete speech signal having the sampling period converted by the sampling period converting means A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 112. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. From the time series of the generated synthesis parameter frame, the first frame period, the second sampling period different from the first sampling period, and the third sampling period different from the second sampling period.
A discrete speech signal is synthesized at a second frame period determined based on the sampling period of, and the sampling period of the discrete speech signal synthesized at the second frame period is converted to the second sampling period. The voice synthesis method, comprising: converting the discrete voice signal after the conversion of the sample period to the sample period into the analog voice signal at the third sample period. 113. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A speech unit accumulating unit for accumulating a plurality of speech units, a voice quality selecting unit for selectively designating a voice quality of a voice to be synthesized, and selecting the speech unit from the speech unit accumulating unit based on input phoneme information. A synthesized parameter frame time series generation means for connecting and generating a time series of synthesized parameter frames, and a time series of synthesized parameter frames generated by the synthesized parameter frame time series generation means are selected and designated by the voice quality selection means. Synthesizing means for synthesizing the discrete speech signal in the second frame period determined according to the voice quality, and the discrete sound synthesized by the synthesizing means. Sampling cycle conversion means for converting the sampling cycle of the signal into a second sampling cycle determined according to the voice quality selected and designated by the voice quality selecting means, and a discrete speech signal having a sampling cycle converted by the sampling cycle conversion means. Different from the second sampling period (except when the second sampling period matches the first sampling period)
A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 114. A discrete speech signal sampled at a first sampling period is cut out in a predetermined synthesis unit from a time series of a characteristic parameter frame of speech obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. In a voice synthesis method for synthesizing a discrete voice signal from a time series of generated synthesis parameter frames, a selection process of a voice quality of a voice to be synthesized is accepted, and a synthesis process for synthesizing a discrete voice signal from the time series of the synthesis parameter frame is performed. , The second frame period determined according to the accepted voice quality, and the sampling period of the synthesized discrete speech signal is changed to the accepted voice quality. According to the second sampling cycle, and the discrete speech signal after the sampling cycle conversion into the second sampling cycle is different from the second sampling cycle (where the second sampling cycle is the first sampling cycle). (Excluding the case where the sampling period is equal to the sampling period) is converted into an analog speech signal at the third sampling period. 115. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A speech unit accumulating means for accumulating a plurality of speech units, and a synthesis parameter frame for selecting and connecting the speech units based on the input phoneme information from the speech unit accumulation means to generate a time series of synthesis parameter frames From the sequence generation means and the time series of the synthesis parameter frames generated by the synthesis parameter frame time series generation means, the first
Second sampling period different from the second sampling period and a third sampling period different from the second sampling period, the second frame period = the first frame period × the second sampling period / the third sampling period. A synthesizing means for synthesizing the discrete speech signal in the second frame cycle defined by the sampling period of And a digital / analog converting means for converting the discrete voice signal whose sample period is converted by the sample period converting means into an analog voice signal at the third sample period. 116. A discrete speech signal sampled at a first sampling period is extracted in a predetermined synthesis unit from a time series of speech characteristic parameter frames obtained by applying a time window of a first frame period to analysis. A plurality of voice units are stored in the voice unit storage unit, and the voice units are selected from the voice unit storage unit based on the input phoneme information and connected to generate a time series of synthesis parameter frames. From the time series of the generated combined parameter frame, the second frame period = the second frame period = the second sample period different from the first sample period and the third sample period different from the second sample period. 1 frame period × second sample period / third sample period, the discrete voice signal is synthesized in a second frame period, and the sample period of the discrete voice signal synthesized in this second frame period is Speech synthesis method characterized by the converted into sample periods, to convert the discrete audio signal after the sample periods conversion to the second sample period into an analog audio signal at the third sampling period. 117. A speech unit accumulating means for accumulating a plurality of speech units created from discrete speech signals sampled at a first sampling period by inputting prosody information and phonological information, and a fundamental frequency of speech from the prosody information. A pitch pattern generating means for generating a pattern; a phonological parameter generating means for generating a phonological parameter of a voice by selectively reading out and connecting a speech element from the speech element accumulating means based on the phonological information; For synthesizing a discrete speech signal from the phoneme parameters generated by the parameter generating means and the fundamental frequency pattern generated by the pitch pattern generating means, and for converting the sampling period of the discrete speech signal synthesized by this synthesizing means And the first sampling period or the first sampling period. Become the second
Sampling period conversion means capable of selectively performing sampling period conversion to the sampling period of, and a discrete speech signal whose sampling period has been converted by the sampling period conversion means, which is different from the second sampling period. And a digital / analog conversion means for converting into an analog audio signal at a sampling period of, the pitch pattern generating means when the sampling period converting means converts the sampling period into the second sampling period. Is configured to generate the fundamental frequency pattern so as to have a fundamental frequency pattern different from that when the sampling period conversion to the first sampling period is performed. 118. A plurality of speech units created from discrete speech signals sampled at a first sampling period with prosody information and phonological information as inputs are accumulated in a speech unit accumulating unit, and the prosody information is used to generate speech. A basic frequency pattern is generated, and a phoneme parameter of the voice is generated by selectively reading out and connecting the phoneme unit from the phoneme unit accumulating unit based on the phoneme information, and the generated phoneme parameter and the basic frequency pattern. In a speech synthesis method for synthesizing a discrete speech signal from a speech signal and converting the synthesized discrete speech signal to an analog speech signal, a first mode in which the sampling period of the synthesized discrete speech signal remains the first sampling period And a second mode in which a second sampling period different from the first sampling period is prepared, and in the second mode, in the first mode, Generate the fundamental frequency patterns so as to have different fundamental frequency patterns and use the synthesized discrete speech signals for synthesis, convert the sampling period of the synthesized discrete speech signals into the second sampling period, A voice synthesizing method, characterized in that a discrete voice signal after conversion of a sample period into a period is converted into an analog voice signal at a third sample period different from the second sample period. 119. A speech segment accumulating means for accumulating a plurality of speech segments created from discrete speech signals sampled at a first sampling period with prosody information and phonological information as inputs, and the first sampling period. Pitch pattern generating means for generating a fundamental frequency pattern of speech from the prosody information based on a different second sampling cycle and a third sampling cycle different from the second sampling cycle; and the speech based on the phonological information. A phoneme parameter generating means for generating a phoneme parameter of the voice by selectively reading out and connecting the phoneme pieces from the phoneme accumulating means, a phoneme parameter generated by the phoneme parameter generating means and the pitch pattern generating means. A synthesizing means for synthesizing the discrete speech signal from the fundamental frequency pattern, and the discrete speech signal synthesized by this synthesizing means. Sampling period conversion means for converting a sampling period into the second sampling period, and digital / analog for converting the discrete speech signal whose sampling period is converted by the sampling period conversion means into an analog speech signal in the third sampling period. A speech synthesis apparatus comprising: a conversion unit. 120. A plurality of speech units created from a discrete speech signal sampled at a first sampling period by inputting prosodic information and phonological information are accumulated in a speech unit accumulating unit, and the first sampling period is stored. Generate a fundamental frequency pattern of speech from the prosody information based on a second sampling period different from the above and a third sampling period different from the second sampling period, and based on the phonological information, the speech unit. The phoneme parameters of the voice are generated by selectively reading and connecting the voice units from the storage means, and the discrete voice signal is synthesized from the generated phoneme parameter and the fundamental frequency pattern, and the sampling period of the synthesized discrete voice signal is It is characterized in that it is converted into the second sampling cycle, and the discrete speech signal after the sampling cycle conversion into the second sampling cycle is converted into an analog speech signal in the third sampling cycle. Speech synthesis method to be. 121. A speech unit accumulating unit for accumulating a plurality of speech units created from a discrete speech signal sampled at a first sampling period by inputting prosodic information and phonological information, and selecting and designating a voice quality of speech to be synthesized. Voice quality selection means, a pitch pattern generation means for generating a fundamental frequency pattern of a voice from the prosody information according to the voice quality selected and designated by the voice quality selection means, and the voice segment accumulation based on the phonological information. A phoneme parameter generating means for generating a phoneme parameter of the voice by selectively reading out and connecting a voice unit from the means, a phoneme parameter generated by the phoneme parameter generating means, and a fundamental frequency generated by the pitch pattern generating means. A synthesizing means for synthesizing a discrete speech signal from a pattern, and a discrete speech synthesized by this synthesizing means Sampling period converting means for converting the sampling period of the signal into a second sampling period determined according to the voice quality selected and designated by the voice quality selecting means, and a discrete speech signal whose sampling period is converted by the sampling period converting means. Different from the second sampling period (except when the second sampling period matches the first sampling period)
A voice synthesizer, comprising: a digital / analog conversion means for converting into an analog voice signal at a third sampling period. 122. A plurality of speech units created from a discrete speech signal sampled at a first sampling period by inputting prosody information and phonological information are accumulated in a speech unit accumulating means, and the speech is converted from the prosody information. A basic frequency pattern is generated, and a phoneme parameter of the voice is generated by selectively reading out and connecting the phoneme unit from the phoneme unit accumulating unit based on the phoneme information, and the generated phoneme parameter and the basic frequency pattern. In a speech synthesis method for synthesizing a discrete speech signal from, the selection and designation of the voice quality of the speech to be synthesized is accepted, the fundamental frequency pattern used in the synthesis process is generated from the prosody information according to the accepted voice quality, and the synthesis is performed. The sample period of the discrete voice signal is converted into a second sample period determined according to the received voice quality, and the second sample period Of the discrete speech signal after the conversion of the sampling period into the period is different from the second sampling period (except when the second sampling period matches the first sampling period) A speech synthesis method characterized by converting into a signal. 123. The pitch pattern generation means performs the sampling cycle conversion to the first sampling cycle when the sampling cycle conversion means performs the sampling cycle conversion to the second sampling cycle. The speech synthesizer according to claim 117, wherein a fundamental frequency pattern whose fundamental frequency is (third sampling period / second sampling period) times that of the fundamental frequency pattern generated when the speech synthesis is performed is generated. 124. The fundamental frequency pattern used in the synthesis process in the second mode has a fundamental frequency of the fundamental frequency of the fundamental frequency pattern used in the synthesis process in the first mode (third sampling period). 120./The second sampling period). 125. Accumulating means for accumulating voice characteristic parameters, synthesizing means for synthesizing a discrete speech signal with the speech characteristic parameters read from the accumulating means as an input, and the discrete speech signal synthesized by this synthesizing means. A digital / analog converting means for converting into an analog audio signal, a sampling period of the discrete audio signal inputted to the digital / analog converting means, and a digital / analog converting means for converting the discrete audio signal into an analog audio signal. A speech synthesizer comprising: a control unit that controls the conversion period so that it is different from the conversion period. 126. Characteristic parameters of speech are accumulated in accumulating means, discrete characteristic speech signals are synthesized using the characteristic parameters of speech read out from the accumulating means as inputs, and the synthesized discrete speech signals are converted into analog speech signals. In the speech synthesizing method, the sampling period of the discrete voice signal to be converted into the analog voice signal is different from the conversion period when the discrete voice signal is converted into the analog voice signal. Speech synthesis method. 127. Accumulating means for accumulating voice characteristic parameters, synthesizing means for synthesizing discrete speech signals with the speech characteristic parameters read from the accumulating means as inputs, and the discrete speech signals synthesized by the synthesizing means. Digital / analog conversion means for converting into an analog voice signal, voice quality selection means for selectively designating voice quality of voice to be synthesized, sampling period of discrete voice signal inputted to the digital / analog conversion means and the discrete voice signal And a control means for varying a ratio with a conversion period when the digital-to-analog converting means converts into an analog voice signal according to the voice quality selected and designated by the voice quality selecting means. Synthesizer. 128. A voice characteristic parameter is accumulated in a storage means, a voice characteristic parameter read from the storage means is input to synthesize a discrete voice signal, and the synthesized discrete voice signal is converted into an analog voice signal. In the speech synthesis method described above, the selection and designation of the voice quality of the speech to be synthesized is accepted, and the sampling period of the discrete speech signal to be converted into the analog speech signal and the conversion of the discrete speech signal into the analog speech signal are performed. A voice synthesizing method, characterized in that a ratio to a conversion cycle is varied according to the received voice quality. 129. Storage means for accumulating pre-created speech units, and synthesis means for synthesizing discrete speech signals by selecting and connecting the speech units from the storage means based on given phoneme information. A digital / analog converting means for converting the discrete audio signal synthesized by the synthesizing means into an analog audio signal, a sampling period of the discrete audio signal inputted to the digital / analog converting means, and the discrete audio signal for the digital / analog converting means. A voice synthesizer, comprising: a control unit that controls so that a conversion cycle when the analog conversion unit converts the analog voice signal is different. 130. A pre-created speech unit is stored in a storage unit, and a speech unit is selected from the storage unit based on given phoneme information and connected to synthesize a discrete speech signal, In a voice synthesis method for converting the synthesized discrete voice signal into an analog voice signal, a sampling period of the discrete voice signal to be converted into the analog voice signal and a conversion for converting the discrete voice signal into the analog voice signal A speech synthesis method characterized in that the period is different. 131. An accumulating means for accumulating a speech element prepared in advance, and a synthesizing element for synthesizing a discrete speech signal by selecting and connecting the speech element from the accumulating element on the basis of given phoneme information. A digital / analog converting means for converting the discrete voice signal synthesized by the synthesizing means into an analog voice signal, a voice quality selecting means for selectively designating a voice quality of voice to be synthesized, and a digital / analog converting means for inputting the voice quality. The ratio of the sampling period of the discrete voice signal to the conversion period when the discrete voice signal is converted into the analog voice signal by the digital / analog converting means is changed according to the voice quality selected and designated by the voice quality selecting means. A voice synthesizer comprising: a control unit. 132. Preliminarily produced speech units are stored in a storage unit, and the speech units are selected and connected from the storage unit based on given phoneme information to synthesize a discrete speech signal, In the voice synthesis method for converting the synthesized discrete voice signal into an analog voice signal, the selection and designation of the voice quality of the voice to be synthesized is accepted, and the sampling period of the discrete voice signal to be converted into the analog voice signal and A voice synthesizing method, characterized in that a ratio of a conversion period when a discrete voice signal is converted into an analog voice signal is varied according to the received voice quality.