JPH1195797A - Device and method for voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a voice synthesizer that can unprove the quality of synthesized voices by employing an extremely easy means without recording uttered voices of an announcer and resegmenting voice elements. SOLUTION: The synthesizer is provided with a voice element memory 19, which accumulates the voice elements of the voices obtained by analyzing the discrete voice signals that are sampled in a first sampling period, a sampling frequency conversion processing section 33, which selectively reads the voice elements from the memory 19 based on given phoneme information and couverts the sampling period of the read voice elements into a second period that differs from the first period, and a waveform superimposing process section 20 which successively connects the sampling period of the voice elements, that is converted into the second period by the section 33, and synthesizes voices.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to voice synthesis software or a voice synthesizer for reading text (sentences mixed with kanji and kana) by voice, and more particularly to voice quality conversion for diversifying synthesized voice.

[0002]

2. Description of the Related Art Conventionally, it is known that there is a voice synthesizer capable of subdividing and storing a voice waveform and synthesizing voice by a combination thereof as a voice synthesizer. Hereinafter, a conventional technique of the speech synthesizer will be described with reference to the drawings. FIG. 6 is a diagram showing a configuration of a conventional speech synthesizer.

The speech synthesizer shown in FIG. 6 converts a text into a symbol string composed of phonemes and prosody, and generates a speech from the symbol string (Text-To-Speech conversion; hereinafter T).
TS). The TTS process in the speech synthesizer in FIG. 6 is roughly divided into two processing units, a language processing unit 101 and a speech synthesis unit 102. Taking the rule synthesis of Japanese as an example, the TTS process is performed as follows. General.

First, a language processing unit 101 performs language processing such as morphological analysis and syntax analysis on text (kanji mixed sentence) input from a text file 103 to decompose it into morphemes, estimate dependency relations, and the like. At the same time, the pronunciation and accent type are given to each morpheme.
After that, the language processing unit 101 determines an accent type for each phrase (hereinafter referred to as an accent phrase) serving as a delimiter at the time of reading aloud using accent movement rules such as a compound word for the accent. Normally, in the language processing unit of TTS,
The reading and accent type for each accent phrase obtained in this way can be output as a symbol string (hereinafter referred to as a phonetic symbol string).

Next, in the speech synthesis unit 102, the duration of each phoneme included in the obtained reading is sent to the phoneme duration calculation processing unit 104 according to a predetermined rule based on the phoneme environment of the phoneme and the like. To decide. Further, the pitch pattern generation processing unit 105 sets a point pitch at a point in time when a change in pitch occurs based on the accent type included in the phonetic symbol string, and performs linear interpolation between the plurality of set point pitches to perform pitch interpolation. Is generated, and an intonation component (usually a monotonically decreasing straight line on the frequency axis) is superimposed on the accent component to generate a pitch pattern.

The speech unit memory 107 in the figure is composed of a large number of speech units created in advance. The speech unit is obtained by cutting out all syllables included in the Japanese speech from a speech waveform uttered by an announcer or the like in a predetermined synthesis unit, for example, a Japanese syllable (consonant + vowel: hereinafter referred to as CV). Created. Further, in order to change the pitch of the voice at the time of voice synthesis, the voice segment cut out as described above is further divided into pitch units in the voiced section of the voice. How to divide into pitch units
As shown in FIG. 7, this is generally performed by applying a window function (here, a Hanning window) of about 1 to 2 times the pitch period.

[0007] The speech synthesis processing unit 107, based on the "reading" included in the speech symbol string, the speech unit memory 107
, Necessary speech units are sequentially read out. Then, in accordance with the determined “length of phoneme duration”, the pitch of the speech unit is increased in the unvoiced section in a period based on the pitch pattern already generated in the voiced section while expanding and contracting the stored waveform in the time axis direction. A desired speech is synthesized by superimposing and repeating unit waveforms.

The processing up to this point is generally performed by a program. Therefore, the synthesized speech is a discrete signal. Therefore, the speech synthesizing unit 102 finally supplies the discrete signal to the D / A converter 108. I do. D
The / A converter 108 converts the discrete signal into an electric analog signal (analog audio signal). By driving the loudspeaker 110 and the like via the amplifier 109 with the analog signal thus obtained, a sound that can be perceived by hearing can be synthesized.

[0009]

The present invention relates to a speech synthesizer,
At present, the above-mentioned conventional techniques exist, but the speech synthesized by the speech synthesizer of this technique has the following problems. That is, in the conventional speech synthesizer, the type of synthesized speech voice (hereinafter referred to as "voice quality") is limited, and the voice quality of the announcer at the time of creating the speech unit file, or the voice quality is slightly degraded due to the rule synthesis of voice. Can only be synthesized with sound quality.

Therefore, when attempting to increase the voice quality of synthesized speech in speech synthesis of a conversational sentence or the like, the speech synthesizer developer newly hires a different announcer, records the utterance, and prepares a speech unit. You have to start over. For this reason, a wage for hiring the announcer is required, and the developer requires a great deal of labor to record the utterance of the announcer and cut out the speech unit.
This leads to an increase in device development costs.

The present invention has been made in view of such circumstances, and has as its object to record the voice quality of synthesized speech by extremely easy means without recording announcer utterances or re-cutting out speech units. It is an object of the present invention to provide a speech synthesis device and method capable of increasing the number of speech synthesis.

[0012]

Means for Solving the Problems To solve the above problems, a storage means for storing characteristic parameters of a voice obtained by analyzing a discrete voice signal sampled at a first sampling period, and a given phoneme. Sample cycle conversion means for selectively reading the feature parameter from the storage means based on information, and converting a sample cycle of the read feature parameter into a second cycle different from the first cycle;
And a speech synthesizing unit for sequentially connecting the sample periods of the characteristic parameter converted into the second period by the sample period converting unit to synthesize a voice.

According to the configuration of the present invention, since the speech spectrum is shifted on the frequency logarithmic axis by making the first cycle and the second cycle different, speech having different voice qualities can be produced using the same speech parameter. Can be synthesized.

Further, in the present invention, the sampling period conversion means converts the sampling period into a second period different from the first period only in a voiced section of the voice. According to the configuration of the present invention, since the spectrum of the voice is shifted on the frequency logarithmic axis by making the first cycle and the second cycle different, voices of different voice qualities are synthesized using the same voice parameter. And the spectrum of the unvoiced section does not change since it operates only when the voice is in the voiced section, so that the clarity is not lost in the unvoiced section of the synthesized voice.

Further, in the present invention, the sampling period conversion means has a switching means for converting the sampling period of the voice read from the storage means into a second period different from the first period. .

According to the structure of the present invention, since the spectrum of the voice is shifted on the frequency logarithmic axis by making the first cycle and the second cycle different, voices having different voice qualities using the same voice parameter are used. Can be synthesized,
Further, it is possible not to operate in a voice section in which a spectrum change adversely affects the synthesized voice.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of a speech rule synthesizing apparatus according to a first embodiment of the present invention.

The speech rule synthesizing apparatus shown in FIG. 1 is realized by executing dedicated software (sentence / speech conversion software) on an information processing apparatus such as a personal computer, and has a sentence / speech conversion (TTS) processing function. That is, it has a sentence-to-speech conversion process (sentence-to-speech synthesis process) function of generating speech from text, and its functional configuration is roughly divided into a language processing unit 1 and a speech synthesis unit 3.

The language processing unit 1 analyzes input sentences, for example, sentences mixed with kanji and kana, to generate reading information and accent information, and based on the information, a phonetic symbol sequence and a speech symbol string in which accent information is described. Controls the process of generating The speech synthesis unit 3 controls a process of generating a speech based on a speech symbol string output from the language processing unit 1.

Now, in the speech rule synthesizer of FIG.
A document (here, a Japanese document) to be subjected to sentence-to-speech conversion (speech) is stored as a text file 5. In this device, according to the sentence-to-speech conversion software, a sentence mixed with kanji or kana is read from the file one by one,
The sentence-to-speech conversion process described below is performed by the language processing unit 1 and the speech synthesis unit 3 to synthesize speech.

First, the sentence mixed with kanji and kana read from the text file 5 is input to the language analysis processing unit 7 in the language processing unit 1. The linguistic analysis processing unit 7 performs a morphological analysis on the input kanji-kana mixed sentence to generate reading information and accent information. Morphological analysis is an operation to revise which character string forms a phrase in a given sentence and what the structure of the word is.

For this purpose, the linguistic analysis processing unit 7 uses a morphological dictionary 9 having "morpheme", which is the minimum component of a sentence, as a headword and a connection rule file 11 in which connection rules between morphemes are registered. That is, the linguistic analysis processing unit 7 obtains all morpheme sequence candidates obtained by matching the input sentence with the morphological dictionary 9 (brute-force method), and grammatically refers to the connection rule file 11 from among them. Output combinations that can be connected before and after.

In the morphological dictionary 9, grammatical readings and accent types are registered together with grammatical information used at the time of 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, "Go to the park and read a book."
When you perform a morphological analysis on the sentence "/ Park / Go / Go / Book / Read / Read /".

And morphemes. Yomi and accent type are given to each morpheme, and they are / Kouen / D / Itte / Phon / ヲ / Yomi / Mas /. Here, the morpheme containing “＾” means that the pitch is high in the syllable immediately before the syllable, and the pitch is decreased in the syllable immediately after the syllable. When there is no "＾", it means that the accent is a flat type.

By the way, when a human reads a sentence, he does not read with such an accent in morpheme units, but reads several morphemes together and adds an accent to the group.

In view of the above, the linguistic analysis processing unit 7 further summarizes the morphemes by one accent phrase (accent giving unit), and also estimates the movement of the accent due to the summary. In addition to this, the linguistic analysis processing unit 7 also adds information such as vowel devoicing and pause (breathing) at the time of reading out. In the above example, the following reading information is finally generated.

/ Cohenue / Itte. Here, the period “.” Represents a breath, and “()” represents a syllable in which a vowel is unvoiced.

When the reading information is generated by the language analysis processing section 7 in the language processing section 1 as described above, the phoneme duration calculation processing section 13 in the speech synthesis section 3 is activated. The phoneme duration calculation unit 13 calculates the duration 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 7 (unit is m
s) is determined.

The processing for determining the duration in the phoneme duration calculation processing unit 13 is performed such that the positions of the boundaries (crossing the CV) between the consonants (C) and the vowels (V) are arranged at equal intervals. It is realized by a simple algorithm.
The interval over the CV is given by the speech rate control unit 15. Although not shown, the software used in the present embodiment allows the user to specify the speed of the synthesized speech. Then, the speech speed specified by the user is given to the utterance speed control unit 15, so that the utterance speed control unit 15 is determined by the duration determination processing in the phoneme duration calculation unit 13. The speed of the synthesized voice is actually changed by adjusting the interval between the CVs. However, it is known from analysis results that the duration of consonants is almost constant even if the speed of utterance is changed, so the duration of consonants is kept constant, and the duration of vowels is kept constant. To change the interval across the CV.

When the duration of each syllable (consonant part and vowel part) included in the input sentence is determined by the phoneme duration calculation processing unit 13, the pitch generation processing unit 17 in the same speech synthesis unit 3 Is activated. The pitch generation processing unit 17 first determines the point pitch based on the duration determined by the phoneme duration calculation processing unit 13 and the accent information determined by the language processing analysis processing unit 7 (in the language processing unit 1). Set the position. Next, a plurality of set point pitches are interpolated by a straight line to obtain a pitch pattern every 10 msec, for example.

Next, in this embodiment, speech units are created in advance in the following procedure. First, a range from which a predetermined synthesis unit included in the real voice is cut out from the real voice sampled at the sampling frequency of 11025 Hz is determined. Then, in the unvoiced section within the cutout range, the waveform is stored as it is, and in the voiced section, the pitch waveform repetition waveform is multiplied by a Hanning window function having a window width twice as large as the pitch cycle to obtain all the pitch waveforms included in the voiced section. Extract and save.

As can be seen from this, one synthesis unit (speech unit) is composed of an unvoiced section waveform and a plurality of pitch waveforms in a voiced section (some synthesis units may not include unvoiced sections). Here, a consonant + vowel (CV) is used as the above-mentioned predetermined synthesis unit, and a total of 137 speech units obtained by cutting out all syllables necessary for the synthesis of Japanese speech are prepared and stored in advance. .

However, the voice waveform is changed (in units of pitch waveform).
If you store it as it is, you will need a large amount of memory,
It is common to store them in compressed form. In the present embodiment, a linear prediction coefficient obtained by performing linear prediction analysis on a pitch waveform and a linear prediction residual signal obtained by passing a speech waveform through a linear prediction inverse filter composed of the obtained linear prediction coefficients, All pitch waveforms included in the 137 speech units are obtained, and all of the obtained linear prediction coefficient / residual waveform pairs are compressed by vector quantization which is often used in the field of speech compression coding. It is carried out. The compressed speech segments are stored in a speech segment file (not shown) prepared in advance.

The contents of the speech unit file are obtained by regenerating a linear prediction coefficient and a residual signal pair from the speech unit compressed as described above at the start of the sentence-to-speech conversion process according to the sentence-to-speech conversion software. All original pitch waveforms were resynthesized (ie, decoded) by a linear prediction filter
Thereafter, the data is read into, for example, a speech unit area (hereinafter referred to as a speech unit memory) 19 secured in a main memory (not shown).

The voice synthesizing unit 21 stores the voice unit memory 1
9 includes a waveform superimposition processing unit 20 for superimposing a pitch waveform of a speech unit read from 9 and a sampling frequency conversion processing unit 33 for converting a sampling frequency of a speech unit read from the speech unit memory 19. Become.

Further, the sampling frequency conversion processing section 33 reads the pitch waveform from the speech unit memory 19 and converts the sampling frequency of the pitch waveform.
It has a switching circuit 35 for not performing conversion, and in accordance with a command from the voice quality control unit 31 (described later), the switching circuit 3
5 is controlled.

Here, a case where the sampling frequency of the speech unit is not converted will be described. First, when the sampling frequency conversion control unit 37 obtains from the voice quality control unit 31 information that the sampling frequency is not to be converted, the switch of the switching circuit 35 is switched to the switch 1 (S1).
The (original) speech unit is read from the speech unit memory 19.

In this way, the waveform superposition processing section 20 determines the read speech unit by the phoneme duration calculation processing section 13 according to the phoneme information in the speech symbol string passed from the language processing section 1. In accordance with the duration, linear expansion and contraction are performed in the unvoiced section, and the pitch waveform is repeated or thinned out in the voiced section to synthesize a speech signal having a desired phoneme duration. In the voiced section, a plurality of pitch waveforms included in the read speech unit are superimposed at a cycle calculated from the pitch pattern generated by the pitch generation processing unit 17, thereby producing a voice of a desired pitch. Are synthesized.

As described above, the voice synthesized by the voice synthesis processor 21 is a discrete voice signal, which is converted into an analog signal by the D / A converter 23 and output to the speaker 27 through the amplifier 25. You can hear it as a sound for the first time.

The point of the present embodiment is that a voice quality switching unit 29 and a voice quality control unit 31 are added to the voice synthesis unit 3, and before the voice synthesis processing unit 21 superimposes the pitch waveform, the sampling frequency conversion processing unit 33. Is that the sampling frequency of the pitch waveform is converted.

The voice quality switching section 29 can specify the voice quality at the time of synthesis by an application program or the like using a user or a sentence-to-speech conversion software. In the present embodiment, it is assumed that three types of voice qualities can be designated by the voice quality switching unit 29.

The voice quality control unit 31 sets the sampling frequency of each pitch waveform in the voiced section of the voice unit read from the voice unit memory 19 to 11025 Hz in accordance with the voice quality specified by the voice quality switching unit 29. (No conversion) The sampling frequency conversion processing unit 33 is controlled so as to convert to any of 12000 Hz and 10000 Hz.

Here, the sampling frequency conversion processing unit 3
3 will be described in detail. Various methods can be applied to the sampling frequency conversion. In this embodiment, it is assumed that a simple method having the configuration shown in FIG. 2 is used.

The sampling frequency conversion processing section 33 is shown in FIG.
As shown in (a), the sampling frequency expander 51,
It comprises a low-pass filter (LPF) 52 and a sampling frequency compressor 53.

In the sampling frequency conversion processing section 33 shown in FIG. 2A, from the sampling frequency f1 to f2 = (L /
To convert the frequency to M) f1, first, as shown in FIG. 2B, the sampling frequency expander 51 inserts (L-1) zero samples s0 between the samples s1. Next, the audio data output from the sampling frequency expander 51, that is, the audio data in which (L-1) zero samples s0 are inserted between the samples s1, is reduced to f1 or f2 to prevent aliasing. (Min
(F1, f2)) is passed through a low-pass filter (low-pass digital filter) 52 having a cutoff frequency. Here, in order to prevent a decrease in gain (1 / L times) due to insertion of a zero sample in the sampling frequency expander 51, the low-pass filter 52 has a gain of L times in the pass band. Finally, the sampling frequency compressor 53 converts the audio data that has passed through the low-pass
As shown in (b), by performing a thinning process of extracting only one sample for every M samples, audio data of a sampling frequency f2 = (L / M) f1 is obtained. FIG. 2B shows a simple example of converting the sampling frequency f1 to f2 = (4/3) f1 for easy understanding.

Here, as described above, if the sampling frequency is converted from 11025 Hz to 12000 Hz, f1 = 111025 [Hz] f2 = 12000 [Hz] f2 = (12000/11025) f1 = (160/147) ) F1, the sampling frequency conversion processing unit 33 calculates L = 160 M = 147 (cutoff frequency of LPF) = min (f1, f2) = f1
= 111025 [Hz] and the above processing may be performed.

Similarly, the sampling frequency is set to 11025
In the case of converting from Hz to 10000 Hz, since f1 = 111025 [Hz] f2 = 10000 [Hz] f2 = (10000/11025) f1 = (400/441) f1, the sampling frequency conversion processing unit 33 L = 400 M = 441 (cutoff frequency of LPF) = min (f1, f2) = f1
= 10000 [Hz] and the above-described processing may be performed.

In the sampling frequency conversion processing unit 33,
The sampling frequency of the pitch waveform is the same as the sampling frequency when the speech unit was created,
That is, when converting to 11025 Hz (when the sampling frequency is not converted), voice synthesis can be performed using the voice quality of the original voice as described above.

On the other hand, the case of converting to another sampling frequency will be described. The conversion here is 11025
Hz to 12000 Hz, or 11025 Hz to 10,000 Hz.

First, if one of the three voice qualities is specified by the user, the voice quality switching unit 29 transmits to the voice quality control unit 31 that a certain voice quality has been specified. The voice quality control unit 31 controls the sampling frequency conversion processing unit 33 to convert the sampling frequency of the pitch waveform included in the voice segment read from the voice unit memory 19 according to the voice quality. Under the control of the voice quality control unit 31, the sampling frequency conversion processing unit 37 switches the switch of the switching circuit 35 to the switch 2 (S2), and determines the sampling frequency of the pitch waveform included in the speech unit by the method described above. From 11025 Hz to 10,000 Hz. In this case, the number of samples of the pitch waveform whose sampling frequency is changed changes, and in this conversion, the number of samples of the pitch waveform decreases as shown in FIG. (That is, sampling frequency conversion is the same as thinning out and interpolating samples.) Similarly, conversion to a sampling frequency of 12000 Hz is also possible, and a description thereof will be omitted. Although the sampling frequency was changed in this way, 11
When the D / A conversion is performed at 025 Hz, the effect shifted in the frequency axis direction is obtained as shown in FIG.

As shown in FIG. 5, the waveform superposition processing section 20 superposes a plurality of pitch waveforms included in the read speech unit after performing sampling frequency conversion according to the above-described method and superimposing the plurality of pitch waveforms, thereby obtaining a speech having a desired pitch. Can be synthesized. The discrete sound signal obtained in this way is 11025 Hz
The voice quality of the analog voice signal that has been D / A-converted is different from the voice quality of the voice that is the basis of the voice unit.

Further, as described above, the sampling frequency conversion processing is performed only in the voiced section of the voice, and the conversion processing is not performed in the unvoiced section, so that the shift occurs in the unvoiced section spectrum in the frequency axis direction. No decrease in clarity such as unvoiced fricatives occurs.

The embodiments of the present invention have been described above, but the present invention is not limited to these embodiments. For example, in the embodiment described above, the synthesis unit is a syllable, but this is not a limitation. As for speech units, compressed (encoded) speech units are decoded and then read into the speech unit memory. However, they are loaded into the speech unit memory without decoding, and decoded during synthesis. The waveforms may be superimposed, which increases the amount of computation during speech synthesis, but can save memory capacity during execution. Although the number of voice qualities is three in the embodiment, it may be two or four or more. There is no problem even if a syntax analysis or the like other than the morphological analysis is inserted in the language processing unit, and the present invention can be applied not only to the TTS of Japanese but also to the TTS of English and other languages.

Further, regarding the method of determining the duration, the CV
Instead of a method in which the intervals are constant, control based on a statistical method may be used. The present invention can be applied to pitch generation not only by the point pitch method but also by using, for example, a Fujisaki model. in short,
The present invention can be implemented with various modifications without departing from the gist thereof.

[0056]

As described above in detail, according to the present invention, by making the first cycle and the second cycle different, the spectrum of the voice is shifted on the frequency logarithmic axis. Thus, voices having different voice qualities can be synthesized, and the spectrum of the unvoiced section does not change because the voice operates only in the voiced section. Therefore, the clarity is not lost in the unvoiced section of the synthesized voice. In addition, it is possible to prevent the effect from occurring in a speech section in which the spectrum change adversely affects the synthesized speech.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a speech rule synthesis device according to an embodiment of the present invention.

FIG. 2 is a diagram for describing a configuration of a sampling frequency conversion processing unit and its operation according to the embodiment;

FIG. 3 is a diagram showing a pitch waveform width after conversion of a sampling frequency according to the embodiment.

FIG. 4 is a diagram for explaining an effect obtained by changing a sampling frequency of a synthesized voice to be input to a D / A converter according to the embodiment;

FIG. 5 is a diagram showing generation of a synthesized speech waveform by superimposing a plurality of pitch waveforms included in speech segments according to the embodiment.

FIG. 6 is a block diagram showing a schematic configuration of a conventional speech rule synthesis device.

FIG. 7 is a diagram showing a conventional method of dividing a voiced section of a voice into pitch units.

[Explanation of symbols]

1, 101 language processing unit 3, 102 voice synthesis unit 5, 103 text file 7 language analysis processing unit 9 morphological dictionary 11 connection rules 13, 104 ··· Phoneme duration calculation processing unit 15 ··· Speech rate control unit 17 and 105 ··· Pitch generation processing unit 19 and 106 ··· Voice unit memory 20 ··· Waveform superimposition processing unit 21 and 107 ··· Voice synthesizer 23, 108 D / A converter (digital / analog converter) 25, 109 Amplifier 27, 110 Speaker 29 Voice quality switching unit 31 Voice control Unit 33 ... Sampling frequency conversion processing unit 35 ... Switching circuit 37 ... Sampling frequency conversion control unit

Claims (12)

[Claims] An accumulating means for accumulating speech segments generated from discrete speech signals sampled at a first sampling period; and selectively accumulating the speech segments from the accumulating means based on given phoneme information. Reading and converting the read sample period of the speech unit into a second period different from the first period; and a sample period converting unit that converts the sample period into the second period by the sample period conversion unit. A voice synthesizing means for sequentially connecting voice units to synthesize voice. 2. A storage means for storing a plurality of pitch waveforms of a first sample period, a pitch waveform being sequentially read out from said storage means based on given phonological information, and a sample period of the read pitch waveform being stored in a second one. A sample period conversion unit for converting the sample period into a second period different from the first period, and a pitch waveform obtained by converting the sample period by the sample period conversion unit is superimposed on the basis of given prosody information. A voice synthesizing device comprising: a waveform superimposing unit that synthesizes voice. 3. A storage unit for storing a speech unit created from a discrete speech signal sampled at a first sampling period, and the speech unit read from the storage unit based on given phonemic information. Sample period converting means for converting the sample period of the sample period from the first period to a different second period in units of pitch waveforms; A speech synthesizing device, comprising: speech synthesizing means for synthesizing. 4. The speech synthesizer according to claim 1, wherein the sample period conversion means converts the sample period into a second period different from the first period only in a voiced section of the voice. 5. The sampling period conversion unit according to claim 1, further comprising a switching unit configured to convert a sampling period of the voice read from the storage unit into a second period different from the first period. A speech synthesizer as described. 6. A voice quality selecting means for selecting a voice quality of a voice to be synthesized, wherein the sample period converting means converts the voice quality into a second cycle corresponding to the voice quality selected by the voice quality selecting means. Item 4. The speech synthesizer according to any one of Items 1 to 3. 7. A speech unit created from a discrete speech signal sampled at a first sampling period is stored in a storage unit, and the speech unit is selected from the storage unit based on given phoneme information. Converting the sample period of the read speech unit into a second sample period different from the first sample period, and sequentially connecting the converted speech units to synthesize speech. A speech synthesis method characterized by the following. 8. A voice synthesizing method for superposing a plurality of pitch waveforms of a first sample cycle to synthesize voice having desired contents, wherein a sample cycle of the pitch waveform is different from a first cycle.
A voice synthesizing method characterized by superimposing a pitch waveform obtained by converting a sampling period after converting the sampling period into a period. 9. A speech unit created from a discrete speech signal sampled at a first sampling period is stored in a storage unit, and the voice read out from the storage unit based on given phoneme information. A speech synthesis method comprising: converting a sampling period of a unit from the first period to a second period different from each other in units of pitch waveforms; and sequentially connecting the converted pitch waveforms to synthesize a voice. 10. The speech synthesis method according to claim 7, wherein when converting to a second cycle different from the first cycle, the sample cycle is converted only in a voiced section of the speech. . 11. The speech synthesis method according to claim 7, wherein switching is performed to convert a sample period of the sound read from the storage unit to a second period different from the first period. . 12. The speech synthesis method according to claim 7, wherein a voice quality of the voice to be synthesized is selected and converted into a second cycle based on the selected voice quality by a sampling cycle conversion unit.