JPH10301599A - Voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice synthesizer capable of synthesizing a synthetic voice of various voice tones. SOLUTION: This device is provided with a waveform storage part 3 for storing elemental waveforms as a minimum unit of synthesis so that a waveform is one pitch for a voice sound and a waveform is one phoneme for a non-voice sound, a waveform conversion part 2 converting the elemental waveform stored in the waveform storage part 3 and a synthetic part 1 editing the elemental waveform converted by the waveform conversion part 2 and generating a synthetic sound. Since respective elemental waveforms are converted by the waveform conversion part 2 and the synthetic voice is generated by using the converted elemental waveforms, the voice tone of the synthetic sound is independently revised a pitch and an uttering speed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer of a waveform editing system, and more particularly to a speech synthesizer capable of generating synthesized speech having various voice qualities.

[0002]

2. Description of the Related Art The quality of synthesized speech in text-to-speech synthesis is improving year by year. Along with this, users have begun to seek synthesized voices of “various voice qualities”. A human voice is generated by giving various timbres (spectral envelopes) when a sound generated from a sound source such as a vocal cord passes through a vocal tract. Voice quality is determined by this spectral envelope. Therefore, by changing the spectrum envelope, it is possible to change the voice quality.

[0003] In general, speech synthesizers are broadly classified into two types: a waveform editing system and a parameter analysis / synthesis system.

In the parameter analysis / synthesis method, a spectral envelope is modeled and treated as a parameter having a large influence on voice quality such as a formant. Therefore, the voice quality can be easily changed by changing such parameters. The voice quality conversion method in the parameter analysis / synthesis method is described in, for example, JP-A-8-83098, JP-A-8-248994, JP-A-57-105796, JP-A-56-072,500, and JP-A-58 −
It is disclosed in, for example, Japanese Patent No. 168098. At present, however, the parameter analysis / synthesis method is not practical because the sound quality is lower than that of the waveform editing method.

[0005] On the other hand, in a waveform editing method having relatively good sound quality, the spectral envelope is not modeled and is handled in the state of a sound waveform. Therefore, it is not easy to change the voice quality in the waveform editing method.

A method of changing a spectrum envelope in a waveform editing system and a speech synthesis system is disclosed in, for example, Japanese Patent Application Laid-Open No. 8-15290.
No. 0 discloses this. FIG. 20 is a block diagram showing the configuration of the speech synthesizer disclosed in this publication. With reference to FIG.
, An appropriate waveform is selected from the waveform storage unit 903, edited, and a synthesized sound is generated. This synthesized sound is subjected to time expansion / contraction (sampling frequency conversion) in the waveform expansion / contraction unit 902.
By doing so, the spectral envelope of the synthesized sound changes. Thereby, the voice quality is changed. 21A and 21B are examples showing the expansion and contraction of a voiced sound, where FIG. 21A shows a waveform before expansion and contraction, and FIG.
Is the waveform after stretching, (C) is the spectrum envelope before stretching,
(D) is the spectrum envelope after expansion and contraction. The voiced sound is configured as a plurality of one-pitch waveforms arranged as shown in FIG. The interval (distance between centers) is the pitch period (reciprocal of the frequency), the total length is the utterance time, and the one obtained by Fourier transforming the one-pitch waveform is the spectral envelope (C). When the waveform is expanded or contracted over time, its spectrum is also expanded or contracted. For example, in the example of FIG. 21, the spectrum envelope (D) before expansion and contraction is obtained by expanding the waveform.
Compared to the spectral envelope after expansion and contraction (C), the spectrum is degenerated in the low frequency direction. By changing the spectral envelope in this way, the voice quality is changed. However, by performing expansion and contraction, the pitch and utterance time also change. In FIG. 21, when (A) and (B) are compared, the expanded and contracted waveform of (B) has twice the pitch period and the utterance time.

To cope with this, when editing a waveform,
That is, it is necessary to reduce the pitch cycle and the utterance time to half of the normal time at the stage (A) in FIG. In general, when any conversion is performed after waveform editing, it affects not only the spectral envelope but also the pitch and utterance time. When editing the waveform, it is necessary to consider this effect.

[0008]

The technique shown in FIG. 20 requires extra calculations for changing the utterance speed and pitch during waveform editing in accordance with expansion and contraction. The spectral envelope of the synthesized speech is changed according to a uniform rule regardless of the phoneme. Such changes are believed to correspond to the shape of the vocal tract. However, human voice is generated by adding a timbre (spectral envelope) when a sound generated by a sound source such as a vocal cord passes through a vocal tract. Therefore, the spectral envelope is determined by the shape of the vocal tract through which the sound generated by the sound source has passed. However, the position of the sound source differs depending on the phoneme.
For example, in the case of vowels, the sound source is the vocal cords behind the throat,
In the case of the frictional sound "s", the tongue is brought close to the hard upper lid, and this is used as the sound source. In other words, even if the shape of the vocal tract changes in the same way, the distance of the sound generated from the sound source passing through the vocal tract and the like differ, so that the influence of the change in the shape of the vocal tract differs depending on the phoneme. Therefore, the change of the spectral envelope mainly determined by the shape of the vocal tract through which the sound generated from the sound source passes should be different depending on the phoneme. Also, human speech has a plurality of peaks of a spectral envelope called formants. Such formants have a significant effect on voice quality. For this reason, in order to obtain various voice qualities, it is desired that a plurality of formants can be operated. But,
Comparing the spectral envelope of (C) and the spectral envelope of (D) in FIG. 21, the two formants (peaks of the spectral envelope) in (C) are shifted to half frequency positions in (D). doing.

An object of the present invention is to provide a speech synthesizer capable of synthesizing a synthesized speech having various voice qualities.

[0010]

According to a first aspect of the present invention, there is provided a voice synthesizer for storing a unit waveform (voiced: every pitch, unvoiced: each phoneme) as a minimum unit of synthesis, and the waveform. A waveform conversion unit configured to convert the unit waveform stored in the storage unit; and a synthesis unit configured to edit the unit waveform converted by the waveform conversion unit to generate a synthesized sound. Is converted, and the voice quality of the synthesized sound is changed independently of the pitch and the utterance length by generating a synthesized voice using the converted segment waveform.

The second speech synthesizer of the present invention has a waveform storage unit for storing a unit waveform (voiced: for each pitch, unvoiced: for each phoneme) as a minimum unit of synthesis, and stored in the waveform storage unit. A waveform converter for converting a unit waveform and a synthesizing unit for editing the unit waveform converted by the waveform converter to generate a synthesized sound, wherein the waveform converter expands and contracts the unit waveform over time. And a unit that expands and contracts each unit waveform in the waveform expanding and contracting unit, and generates a synthesized speech using the expanded and contracted unit waveform.

A third speech synthesizer according to the present invention comprises a waveform storage unit for storing a unit waveform (voiced: for each pitch, unvoiced: for each phoneme) as a minimum unit of synthesis, and a source stored in the waveform storage unit. A frequency converting unit for converting a segment waveform and a synthesizing unit for editing a segment waveform converted by the waveform converting unit to generate a synthesized sound, wherein the waveform converting unit performs a Fourier transform on the segment waveform A frequency shifter for shifting the frequency of the unit waveform spectrum Fourier-transformed by the frequency converter, a parameter storage for storing parameters necessary for frequency shifting by the frequency shifter, and a frequency shifter for the frequency shifter. It is composed of a frequency inverse transform unit that performs Fourier inverse transform on the obtained unit waveform spectrum, and the frequency shift unit frequency-shifts each unit waveform spectrum to produce a synthesized voice. And changes the.

A fourth speech synthesizer according to the present invention comprises a waveform storage unit for storing a unit waveform (voiced: for each pitch, unvoiced: for each phoneme) as a minimum unit of synthesis, and a source stored in the waveform storage unit. A frequency converting unit for editing a unit waveform converted by the waveform converting unit and generating a synthesized sound by editing the unit waveform converted by the waveform converting unit, wherein the waveform converting unit performs a Laplace conversion of the unit waveform; A frequency shift unit for shifting the frequency of the unit waveform spectrum Laplace transformed by the frequency conversion unit, a parameter storage unit for storing parameters necessary for frequency shifting by the frequency shift unit, and a frequency shift unit for the frequency shift unit. Frequency conversion unit for inverse Laplace transform of the obtained unit waveform spectrum, and the frequency shift unit frequency-shifts each unit waveform spectrum to produce a synthesized voice. And changes the.

A fifth speech synthesizing apparatus according to the present invention comprises a waveform storage unit for storing a segment waveform (voiced: for each pitch, unvoiced: for each phoneme) as a minimum unit of synthesis, and a source stored in the waveform storage unit. A waveform conversion unit configured to convert the one-sided waveform, and a synthesis unit configured to edit the unitary waveform converted by the waveform conversion unit to generate a synthesized sound, wherein the waveform conversion unit is stored in the waveform storage unit. A matrix operation unit that integrates a matrix into a sample value column vector of a waveform, and a parameter storage unit that stores a matrix used in the matrix operation unit. It is characterized in that conversion equivalent to that of the fourth invention is realized.

According to a sixth aspect of the present invention, there is provided a speech synthesizer according to any one of the first to fifth aspects, wherein the unit waveform is divided into several groups, and conversion is performed using different parameters for each group. It is characterized by the following.

A seventh speech synthesizing apparatus according to the present invention is characterized in that, in the speech synthesizing apparatus according to the sixth invention, in particular, the unit waveform is divided into a voiced sound and an unvoiced sound, and only the voiced sound is converted.

An eighth speech synthesizing apparatus according to the present invention is characterized in that the speech synthesizing apparatus according to any one of the first to seventh aspects further comprises a parameter input section and can set parameters arbitrarily.

According to a ninth speech synthesizer of the present invention, in any one of the first to seventh speech synthesizers, the waveform converter converts a unit waveform using parameters specified in the input text. It is characterized by the following.

According to a tenth speech synthesizer of the present invention, the speech synthesizer according to any one of the first to ninth aspects further comprises a waveform storage unit for storing the unit waveform converted by the waveform conversion unit. A synthesized speech is generated using the unit waveform stored in the storage unit.

An eleventh speech synthesizer according to the present invention is characterized in that, in the speech synthesizer according to any one of the first to ninth aspects, the waveform conversion unit changes parameters from prosody information of the synthesis unit. .

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A speech synthesizing apparatus according to an embodiment of the present invention will be described below with reference to the drawings.

[First Embodiment] FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention. Referring to FIG. 1, a first embodiment of the present invention includes a synthesis unit 1 that edits a segment waveform selected from waveform storage unit 3 based on input of text or the like to generate a synthesized sound, and a segment waveform. And a waveform storage unit 3 for storing a waveform for each phoneme and a label of the waveform.

Here, the unit waveform is 1 for a voiced sound.
It shows a waveform for a pitch, and a waveform for one phoneme for an unvoiced sound. The label indicates phoneme information indicating what sound the waveform is, phoneme environment information indicating what text the waveform was obtained from, and the like. When the phoneme is a voiced sound, the phoneme is composed of a plurality of unit waveforms. Therefore, in the present invention, it is necessary to separate the unit sound for each unit waveform. For this reason, the voiced sound waveform is also provided with a label for performing such separation.

FIG. 2 is a block diagram showing the waveform converter 2 of FIG. 1 in more detail. Referring to FIG. 2, waveform conversion unit 2
Is provided with a parameter storage unit 2a that stores expansion and contraction rates, and a waveform expansion and contraction unit 2b that performs time expansion and contraction on the unit waveform using the expansion and contraction rates stored in the parameter storage unit 2a.

Next, the operation of the first embodiment of the present invention will be described with reference to FIGS.

When phonetic information such as text or phonetic symbols to be synthesized is input to the synthesizing unit 1,
The synthesizing unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3.

The waveform storage unit 3 stores, for example, waveforms delimited by phonemes and their labels. The label is given, for example, “koNn; i; chiwa”. Here, "koNn" before the first semicolon (;) is a phoneme chain immediately before the waveform, "i" between two semicolons is "phoneme" of the waveform, and "chiwa" after the second semicolon. Indicates a phoneme chain immediately after the waveform.

When the waveform is a voiced sound,
As described above, it also has a label for separating each unit waveform. This label gives, for example, the start and end times of the unit waveform in the form of "10;21;31;40;...".
In this case, 0 msec to 10 msec of the voiced sound waveform
The part of c is the first one-pitch waveform, 10 msec to 21
msec indicates the next one-pitch waveform...

For selecting a unit waveform, for example, when "a" is necessary for synthesis, a phoneme waveform whose phoneme is "a" is searched for. When there are a plurality of candidates, a search is made for the same phoneme as "a" to be synthesized using the phoneme environment information and the same phonemes before and after. For example, "ka
When it is desired to synthesize “a” of “kigouri”, a waveform having a label “ ^* k; a; k ^* ” indicating that the phoneme before and after a is k is searched. Here, “*” indicates an arbitrary phoneme sequence. If there are a plurality of candidates, the number is increased to two before and after. In addition, priorities are assigned in advance to the phoneme waveforms of the same phoneme, in case that there is no match in either the front or the back or there is no match completely. After selection, the voiced waveform of the voiced sound is divided into unit waveforms using labels.

When the selection is made, the waveform expanding / contracting section 2b reads out the selected segment waveform from the waveform storage section 3 and expands / contracts the time with the expansion / contraction rate stored in the parameter storage section 2a. The time expansion / contraction is performed by, for example, a sampling frequency conversion method. If the original frequency of the waveform is 20 kHz and the expansion / contraction rate is 0.5, the sampling frequency conversion is performed from 20 kHz to 10 kHz.

The expanded and contracted unit waveforms are arranged in the synthesizing unit 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

Specifically, for example, in the case of a voiced sound, FIG.
The unit waveform (waveform for one pitch) (A) stored in the waveform storage unit 3 as shown in FIG.
A unit waveform (B) is generated. Next, the synthesizing unit 1 arranges the expanded and contracted unit waveforms (B) at a desired pitch cycle interval until a desired length is obtained, thereby obtaining a synthesized sound (C). On the other hand, in the case of an unvoiced sound, for example, as shown in FIGS. 16A and 16B, the unit waveform (waveform for one phoneme) is expanded and contracted so that the expanded and contracted waveform has a desired length. I do. As a method of adjusting the length in the case of an unvoiced sound, for example, the following method can be used. As shown in FIG. 16A, the waveform after expansion and contraction (A)
If the length is shorter than the desired length, a part (B) is extracted from the expanded and contracted waveform (A) and connected to obtain a waveform (C) having a desired length. FIG. 16 (b)
When the length of the expanded / contracted waveform (A) is longer than the desired length, a part (D) of the expanded / contracted waveform (A) is deleted to obtain a desired length of the waveform (E). Get)

Here, as a comparative example, Japanese Patent Application Laid-Open No. 8-152
In the conventional example disclosed in Japanese Patent Publication No. 900, a waveform (D) in which elementary waveforms (one-pitch waveform) (A) are arranged as shown in FIG. , And a synthesized voice (C) having the utterance length. However, in this method, as can be seen by comparing (D) and (C), the pitch period and the utterance length change due to expansion and contraction. For this reason, in order to obtain the desired pitch period and generation length at the final output (C) stage, the waveforms are arranged in consideration of the effect of expansion and contraction at the stage of arranging the segment waveforms (A) and forming the waveform (D). There is a need. On the other hand, in the present invention, as shown in FIG. 14, the unit waveform (A) is expanded and contracted in advance, and the expanded and contracted unit waveform (B) is arranged to generate a synthesized sound (C). Therefore, when the synthesized waveforms (C) are formed by arranging the unit waveforms (B), they may be arranged so as to have a desired pitch period and utterance length, and no extra calculation is required.

[Embodiment 2] Embodiment 2 of the present invention
Has basically the same configuration as the first embodiment. The difference is in the configuration and operation of the waveform converter 2.

FIG. 3 is a block diagram showing the waveform converter 2 of FIG. 1 according to the second embodiment in more detail. Referring to FIG. 3, waveform conversion unit 2 includes a frequency conversion unit 2c that performs a Fourier transform on the unit waveform, a frequency shift unit 2d that shifts the frequency of the Fourier-transformed unit waveform spectrum,
A parameter storage unit 2a for storing parameters at the time of frequency shift, and a frequency inverse transform unit 2e for performing Fourier inverse transform of the frequency-shifted unit waveform spectrum are provided.

Next, the operation of the second embodiment of the present invention will be described with reference to FIGS.

When sound information such as text or phonetic symbols to be synthesized is input to the synthesizing unit 1, the synthesizing unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3. The selection method is the same as in the first embodiment. When the selection is made, the frequency conversion unit 2c
Reads out the selected unit waveform f (t), performs a Fourier transform, calculates a unit waveform spectrum S (ω), and passes it to the frequency shift unit 2d. The Fourier transform is represented by Equation 1 below.

[0038]

(Equation 1) Upon receiving the unit waveform spectrum x (ω), the frequency shift unit 2d reads a function w (ω) representing the correspondence of the frequency from the parameter storage unit 2a, and executes the frequency shift. The frequency shift is given by Equation 2 below.

[0039]

(Equation 2) FIG. 17 shows an example of the frequency shift. In FIG.
(B) is the original unit waveform spectrum. If this unit waveform spectrum is frequency-shifted using the function represented by (A) in FIG. 17, for example, when ω = 2, w (2) = 1 from (A) in FIG. w (2) = x
(1), the formant of ω = 1 in FIG. 17B is shifted to ω = 2, and as a result, a spectrum as shown in FIG. 17C is obtained. The frequency-shifted unit waveform spectrum x'S is passed to the frequency inverse transform unit 2e. The frequency inverse transform unit 2e performs an inverse Fourier transform on the frequency-shifted unit waveform spectrum x '. The inverse Fourier transform is represented by the following Equation 3.

[0040]

(Equation 3) The unit waveform f (t) subjected to the inverse Fourier transform is arranged in the synthesizing unit 1 so as to have a desired pitch and utterance speed, and is output as synthesized speech.

Here, in the case of the comparative example, FIG.
Both of the two formants (peaks of the spectrum) of the spectrum before the conversion are located at half frequency positions after the conversion of FIG. As described above, conventionally, only two formants can be changed in the same manner. In contrast,
In the case of the present invention, the two formants of ω = 1, 3 in FIG. 17B are ω = 2, 3.5 in FIG. 17C, and can be changed at different ratios.

Although the above description has been made on the assumption that the waveform takes a continuous value, the present invention can also be realized in a case where the waveform takes a discrete value. In this case, the Fourier transform and the inverse Fourier transform are replaced by a discrete Fourier transform and a discrete inverse Fourier transform, respectively. In this case, the unit waveform before conversion is F [n], the unit waveform spectrum before conversion is X [k], the unit waveform after conversion is F '[n], and the unit waveform spectrum before conversion is X [k]. ′
[K], and a function indicating the frequency shift is W [k],
Equations (1), (2), and (3) are replaced by Equations 4 and 5 below.

[0043]

(Equation 4)

[0044]

(Equation 5) Here, N is the number of samples of the unit waveform, and K is the number of samples of the unit waveform spectrum. The above calculation can be performed by FFT.

The frequency shift is a function y of two variables.
(Ω, l) (in the case of discrete values, Y [k, h]),
Expression 6 below can also be used.

[0046]

(Equation 6) Also, Laplace transform and Laplace inverse transform can be used instead of Fourier transform and Fourier inverse transform. In this case, the unit waveform before conversion is f (t), the unit waveform spectrum before conversion is x (s), and the unit waveform after conversion is f ′.
(T), if the unit waveform spectrum before conversion is x '(s) and the function indicating the frequency shift is w (s), equation (1)
Equations (2) and (3) are replaced by Equation 7 below.

[0047]

(Equation 7) Here, s = δ + jω. In this case, the function W (s) representing the frequency shift is a complex function, and more detailed conversion can be realized.

Similarly, when the waveform takes discrete values, Laplace transform and Laplace inverse transform are replaced with Z transform and inverse Z transform, respectively. In this case, the unit waveform before conversion is F
[N], the unit waveform spectrum before conversion is X [z], the unit waveform after conversion is F '[n], the unit waveform spectrum before conversion is X' [z], and the function indicating the frequency shift is Assuming that W [z], Expressions (1), (2), and (3) are replaced by Expression 8 below.

[0049]

(Equation 8) [Third Embodiment] A third embodiment of the present invention has basically the same configuration as the first embodiment. The difference is in the configuration and operation of the waveform converter 2.

FIG. 4 is a block diagram showing the waveform converter 2 of FIG. 1 according to the third embodiment in more detail. Referring to FIG. 4, waveform converting section 2 includes a matrix calculating section 2 f for integrating a matrix into a sample value sequence spectrum of a unit waveform, and a matrix calculating section 2 f
And a parameter storage unit 2a for storing a matrix to be integrated in the step (a).

Next, the operation of the third embodiment of the present invention will be described with reference to FIGS.

When textual information to be combined, such as text or phonetic symbols, is input to the synthesizing unit 1, the synthesizing unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3. The selection method is the same as in the first embodiment. When the selection is made, the matrix operation unit 2f
Reads out the element A [i, n] of the matrix representing the frequency shift from the parameter storage unit 2a and integrates it into the unit waveform F [n]. The integrated unit waveform F ′ [n ′] is output to the synthesis unit 1
Are arranged so as to have a desired pitch and sound generation speed, and are output as synthesized speech. Frequency-shifted waveform F '
[N '] can be expressed as in the following Expression 9.

[0053]

(Equation 9) By calculating A [n ', n] in advance and storing it in the matrix waveform storage unit, it is possible to reduce the amount of calculation at the time of generating a synthesized sound.

In particular, the resolution of W [k] is determined by K, but since the calculation of the equation is independent of K, the value of K can be made sufficiently large.

[Embodiment 4] Embodiment 4 of the present invention
Has basically the same configuration as that of the first to third embodiments. The difference is that the waveform converter 2 divides the unit waveforms into several groups using the labels and assigns a conversion parameter such as a scaling factor or a frequency shift function to each group. The groups are divided according to, for example, the types of phonemes (vowels, fricatives, plosives, etc.), the positions of unit waveforms in the phonemes (for voiced sounds), and the like.

In the case of the comparative example, conversion (expansion / contraction) is performed after arranging the unit waveforms (A) as shown in FIG. For this reason, for example, it is impossible to set different conversion parameters (expansion and contraction rates) for (C1) and (C2) of the waveform (C). On the other hand, in the present invention, when converting (expanding or contracting) the unit waveform (A), if (B1) and (B2) are generated using different conversion parameters (expansion and contraction rates), (C1) ) And (C2) have different conversion parameters (expansion and contraction rate)
It can be. Therefore, it is possible to realize more varied voice quality as compared with the related art. Further, as described in the section of the problem to be solved by the invention, the influence of the change in the shape of the vocal tract differs depending on the phoneme. By taking into account this effect and setting a conversion parameter (ratio of expansion and contraction) for each phoneme, more natural voice quality conversion becomes possible.

Next, the operation of the fourth embodiment of the present invention will be described with reference to FIGS. In the following, description will be given in accordance with the first embodiment, but it is of course possible to realize the embodiment in accordance with the second and third embodiments.

First, grouping according to the label is defined in advance. Here, for example, the labels are “a” and “i”
The case is divided into two cases: "u", "e", and "o"; Then, the expansion and contraction ratios of the two groups are set to 1.2 and 1.1, and the expansion and contraction ratios are stored in the parameter storage unit 2a. When sound information such as a text to be synthesized or a phonetic symbol is input to the synthesizing unit 1, the synthesizing unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3. The selection method is the same as in the first embodiment. When the selection is performed, the waveform expansion / contraction unit 2 b reads the selected segment waveform from the waveform storage unit 3. The waveform expansion / contraction unit 2b performs grouping based on the read unit waveform labels, and selects the expansion / contraction rate of the group from the parameter storage unit 2a. In the previous example, the label is "a"
In the case of “i”, “u”, “e”, and “o”, the expansion and contraction ratio is 1.2
In other cases, an expansion / contraction ratio of 1.1 is selected. The waveform expansion / contraction unit 2b expands / contracts the unit waveform at the selected expansion / contraction rate. The expanded and contracted unit waveforms are arranged in the synthesizing unit 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

[Embodiment 5] Embodiment 5 of the present invention
Has basically the same configuration as that of the first to third embodiments. The difference is that the waveform conversion unit 2 divides the unit waveform into two groups, a voiced sound and an unvoiced sound, and converts only the voiced sound.

Next, the operation of the fifth embodiment of the present invention will be described with reference to FIGS. In the following, description will be given in accordance with the first embodiment, but it is of course possible to realize the embodiment in accordance with the second and third embodiments.

When pronunciation information such as text or pronunciation symbols to be synthesized is input to the synthesis unit 1, the synthesis unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3. The selection method is the same as in the first embodiment. When the selection is made, the waveform expansion / contraction unit 2b
Reads the selected segment waveform from the waveform storage unit 3.
The waveform expansion / contraction unit 2b determines a voiced sound or an unvoiced sound from the label of the read unit waveform. If it is voiced, the unit waveform is expanded / contracted at the expansion / contraction rate stored in the parameter storage unit 2 a and output to the synthesis unit 1. On the other hand, if you are silent,
The data is output to the synthesizing unit 1 as it is. The unit waveforms are arranged in the synthesizing unit 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

A voiced sound is configured as a plurality of unit waveforms (a waveform for one pitch) arranged as shown in FIG. Therefore, the utterance length of the voiced sound is adjusted by changing the number and interval (pitch cycle). On the other hand, in the case of unvoiced sound, one segment waveform (a waveform for one phoneme)
Consists of Therefore, the utterance length depends only on the length of the segment waveform. Therefore, when the unit waveform is expanded or contracted, the utterance length also changes. In order to keep the utterance length unchanged, for example, the processing shown in FIGS. 16A and 16B is required. However, such connection or deletion of the waveform is likely to cause spectral discontinuity at the connection point, leading to sound quality deterioration. Therefore, in the present invention, sound quality degradation can be suppressed by not performing expansion and contraction of unvoiced sound.

[Sixth Embodiment] FIG. 5 is a block diagram showing a configuration of a sixth embodiment of the present invention. Referring to FIG. 5, a sixth embodiment of the present invention has an input unit 4 that receives an input of a conversion parameter from the outside and designates a conversion parameter to waveform conversion unit 2 in addition to the first to fifth embodiments. doing. Synthesis unit 1, waveform conversion unit 2, and waveform storage unit 3
Operates in the same manner as in the first to fifth embodiments.

The input unit 4 receives the input of the conversion parameter, and transmits the information to the parameter storage unit 2 in the waveform conversion unit 2.
Write to a. Here, one or more conversion parameters can be specified. When a plurality is specified, this conversion parameter corresponds to the unit waveform group in the fourth embodiment. When some or all of the conversion parameters are not specified, the conversion parameters that are not specified are values stored in advance in the parameter storage unit 2a. Thereby, for example, the user can obtain his / her favorite synthesized sound.

Next, the operation of the sixth embodiment of the present invention will be described with reference to FIG. In the following, description will be given in accordance with the second and fourth embodiments, but it is of course possible to realize the present invention in a form according to another embodiment.

Before the synthesis, the expansion / contraction ratio is preliminarily written into the parameter storage unit 2a using the input unit 4. As a method of specifying the expansion ratio, for example, a toolbar can be used. When phonetic information such as a text to be synthesized or phonetic symbols is input to the synthesizing unit 1, the synthesizing unit 1
Selects, from the waveform storage unit 3, a unit waveform necessary for synthesizing the text. The selection is the same as in the first embodiment. When the selection is performed, the waveform conversion unit 2 reads out the selected unit waveform from the waveform storage unit 3 and performs expansion and contraction.
For example, when vowels 1.5 and consonants 1.2 are input to the input unit 4, the label in the waveform storage unit 3 is "a".
The unit waveforms of “i”, “u”, “e”, and “o” have expansion and contraction ratios of 1.
5, others expand and contract at an expansion ratio of 1.2. The expanded and contracted unit waveforms are arranged in the synthesizer 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

[Seventh Embodiment] FIG. 6 is a block diagram showing a configuration of a seventh embodiment of the present invention. Referring to FIG. 6, a seventh embodiment of the present invention has a parameter extracting unit 6 for separating conversion parameter information from information such as input text in addition to the first to fifth embodiments. The synthesizing unit 1, the waveform converting unit 2, and the waveform storing unit 3 operate in the same manner as in the first to fifth embodiments.

The parameter extracting section 6 extracts the conversion parameters described in the input text and writes them into the parameter storage section 2a in the waveform converting section 2. One or more conversion parameters can be described in the input text.
When a plurality of parameters are described, the conversion parameters correspond to the unit waveform groups in the fourth embodiment. If some or all of the conversion parameters are not described, the conversion parameters that are not described are values stored in advance in the parameter storage unit 2a. Thus, the user can obtain his / her favorite synthesized sound without having the input unit 4 described above.

Next, the operation of the seventh embodiment of the present invention will be described with reference to FIG. In the following, description will be given in accordance with the second and fourth embodiments, but it is of course possible to realize the present invention in a form according to another embodiment.

When phonetic information such as text or phonetic symbols to be synthesized is input to the parameter extracting unit 6, the parameter extracting unit 6 separates it into text information and expansion / contraction information. Here, for example, "{v1.1} s
1.2 Please turn right at the next intersection. " Here, “＼” is a special symbol, the alphabet after the label is a label (v: vowel, s: consonant) representing a group of unit waveforms, and the number following is the expansion rate. In this case, the expansion and contraction rate of the vowel is 1.1, and the expansion and contraction rate of the consonant is 1.2. When this text is input, the parameter extraction unit 6 separates the expansion and contraction rate from the text, and stores the expansion and contraction rate 1.1 for vowels and 1.2 for consonants in the parameter storage unit 2a. Please turn right. "

The synthesizing unit 1 selects a unit waveform necessary for synthesizing the text from the waveform storage unit 3. The selection is the same as in the first embodiment. When the selection is performed, the waveform expansion / contraction unit 2b reads the selected segment waveform from the waveform storage unit 3 and performs expansion / contraction at the expansion / contraction rate stored in the parameter storage unit 2a. The expanded and contracted unit waveforms are arranged in the synthesizer 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

[Eighth Embodiment] FIG. 7 is a block diagram showing a configuration of an eighth embodiment of the present invention. Referring to FIG. 7, an eighth embodiment of the present invention has a second waveform storage unit 5 for storing a converted unit waveform and its label in addition to the first to fifth embodiments. Further, similarly to the sixth and seventh embodiments, a configuration having the parameter extracting unit 6 and the input unit 4 is also possible. The configuration in such a case is shown in FIGS.

Next, the operation of the eighth embodiment of the present invention will be described with reference to FIG. In the following, description will be given in accordance with the second and fourth embodiments, but it is of course possible to realize the present invention in a form according to another embodiment.

In the eighth embodiment, at the time when the expansion / contraction ratio is defined (before synthesizing), the waveform converter 2 converts the segment waveform read out from the waveform storage 3, and stores it in the second waveform storage 5. Remember. The second waveform storage unit 5 stores the unit waveform converted by the waveform conversion unit 2. Also, the second
The waveform storage unit 5 stores a label copied from the waveform storage unit 3. At the time of creating a synthesized sound, when pronunciation information such as a text to be synthesized or phonetic symbols is input to the synthesis unit 1, the synthesis unit 1 outputs a segment waveform necessary for synthesizing the text to a second unit. Select from the waveform storage unit 5. The selection is the same as in the first embodiment. The unit waveforms are arranged in the synthesizing unit 1 so as to have a desired pitch and utterance speed, and are output as synthesized speech.

As described above, since the segment waveform is converted in advance, in the seventh embodiment, it is possible to reduce the amount of calculation and to increase the processing speed when generating the synthesized sound.

[Ninth Embodiment] FIG. 10 is a block diagram showing a configuration of a ninth embodiment of the present invention. Referring to FIG. 10, in a ninth embodiment of the present invention, a synthesis unit 1 that edits an appropriate unit waveform based on an input to generate a synthesized sound, and determines a unit waveform using prosodic information. It has a conversion unit 7 for converting with a conversion parameter, and a waveform storage unit 3 for storing a unit waveform before expansion and contraction and its label. Referring to FIG. 13, conversion unit 7 includes a parameter storage unit 2 a that stores one or a plurality of conversion parameters, and a parameter change unit 7.
a waveform conversion unit 7c that performs conversion from the conversion parameters obtained from the parameter a, and a progress storage unit 7b that stores the labels of the unit waveforms that have been converted in the past and passes them to the parameter change unit 7a.
And a parameter changing unit 2e that changes the conversion parameters stored in the parameter storage unit 2a based on the output from the prosody provision unit 1b and the progress storage unit 7b and passes the converted parameters to the waveform conversion unit 7c. The waveform conversion unit 7c has a configuration in which the parameter storage unit is removed from the waveform conversion units shown in FIGS. The synthesizing unit 1 performs a syntax analysis of the input text.
a, a prosody providing unit 1b for providing a prosody to the parsed text, and a waveform editing unit 1c for editing and synthesizing an appropriate unit waveform from the text to which the prosody is provided.
Further, similarly to the sixth and seventh embodiments, a configuration having the input unit 4 and the parameter extracting unit 6 is also possible. The configuration in such a case is shown in FIG. 11 and FIG.

Next, the operation of the ninth embodiment of the present invention will be described with reference to FIGS. In the following, description will be given in accordance with the second and fourth embodiments, but it is of course possible to realize the present invention in a form according to another embodiment.

When a text or a phonetic symbol to be synthesized, such as a phonetic symbol, is input to the synthesizing unit 1, the syntactic analysis unit 1a performs a syntactic analysis on this. Here, for example, assuming that the text “turn right at the next intersection” is input, the parsing unit 1a is called a morpheme such as “next / intersection / turn right / turn / please”. Divide into units. Here, “/” indicates the position of division. Next, the prosody provision unit 1b adds prosody information such as accent, intonation, pause, and rhythm in accordance with the output. In the prosody provision unit 1b, for example, the syntax analysis unit 1a
Output of "next / intersection / right / turn / please", "next's / Kosaten / Migi ni / Maki ni /
Give me the reading and the accent. Here, "'" indicates the position of the accent. The output of the prosody provision unit 1b is sent to the waveform editing unit 1c and the parameter changing unit 7a. The parameter changing unit 7a determines the expansion / contraction ratio of each part of the synthesized speech based on the output. For example, it is assumed that the parameter storage unit 2a previously stores the expansion / contraction ratio for each phoneme as shown in FIG. Further, it is assumed that rules for changing the expansion / contraction ratio are determined to be “to prevent a large difference in expansion / contraction ratio between adjacent phonemes” and “to increase the expansion / contraction ratio of phonemes with accents”. In this case, in the case of the conventional synthesizer that does not use the output of the prosody provision unit 1b, the unit waveform of each phoneme is expanded or contracted at the expansion ratio before the change in FIG. On the other hand, in the present invention, the changes 1 and 2 in FIG.
Change the stretch ratio as shown. In this example, change 1 is
According to the rule "to prevent a large difference in expansion ratio between adjacent phonemes", the expansion ratio of / s / between / usa / is closer to the left and right phonemes / u /, / a /. In the second modification, the expansion / contraction ratio of accented / i / is increased based on the rule that “a phoneme with an accent increases the expansion / contraction ratio”. The underlined portions are the changed portions. Therefore, the parameter changing unit 7a holds the change 2 in FIG. 19 as the final expansion / contraction ratio. Waveform editing unit 1
c selects a unit waveform necessary for synthesizing a text from the waveform storage unit 3. When the unit waveform is selected, the progress storage unit 7b stores the label of the selected unit waveform.
Then, the stored label is output to the parameter changing unit 7a. The parameter change unit 7a grasps which part of FIG. 19 is currently being processed from the stored label, and passes the expansion / contraction ratio of that part to the waveform conversion unit 7c. The waveform conversion unit 7c expands and contracts the unit waveform at the expansion and contraction ratio passed from the parameter changing unit 7a, and passes the resultant to the waveform editing unit 1c. The waveform editing unit 1c rearranges the unit waveforms so as to have a desired pitch and utterance length, and outputs the synthesized sound.

The ninth embodiment differs from the other embodiments in that the same unit waveform is converted with different conversion parameters due to the difference in prosody information, so that the voice quality can be changed in more detail. It is.

[0080]

The speech synthesizer according to the present invention has the following effects.

First, unlike the conventional method, when editing the unit waveform, there is no need to calculate the utterance speed and pitch in accordance with the conversion.

Second, the voice quality can be varied more by shifting the frequency of the unit waveform spectrum.

Third, deterioration of sound quality of synthesized speech can be suppressed by not performing conversion of phonemes such that conversion leads to sound quality deterioration.

Fourth, by determining the conversion parameters according to the types of phonemes and the prosody information, the voice quality can be varied more.

Fifth, since there is a waveform storage unit for storing the pre-converted unit waveform and no conversion is performed at the time of synthesis,
The amount of calculation at the time of synthesis can be minimized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech synthesizer according to Embodiments 1 to 5 of the present invention.

FIG. 2 is a diagram illustrating the first embodiment of the present invention, and is a block diagram illustrating a waveform conversion unit illustrated in FIG. 1;

FIG. 3 is a diagram illustrating a second embodiment of the present invention, and is a block diagram illustrating a waveform conversion unit illustrated in FIG. 1;

FIG. 4 is a diagram illustrating a third embodiment of the present invention, and is a block diagram illustrating a waveform conversion unit illustrated in FIG. 1;

FIG. 5 is a block diagram showing a configuration of a speech synthesis device according to a sixth embodiment of the present invention.

FIG. 6 is a block diagram illustrating a configuration of a speech synthesis device according to a seventh embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of a speech synthesis device according to an eighth embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of another example of the eighth embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of still another example of Embodiment 8 of the present invention.

FIG. 10 is a block diagram showing a configuration of a speech synthesis device according to a ninth embodiment of the present invention.

FIG. 11 is a block diagram showing a configuration of another example of Embodiment 9 of the present invention.

FIG. 12 is a block diagram showing a configuration of still another example of Embodiment 9 of the present invention.

FIG. 13 is a block diagram illustrating a conversion unit illustrated in FIG. 10;

FIG. 14 is a diagram for explaining the operation of the voiced sound wave according to the present invention when expanding and contracting.

FIG. 15 is a diagram for explaining an operation when a voiced sound wave of a comparative example is expanded and contracted.

FIG. 16 is a diagram showing an example of a method of increasing / decreasing the time length of an unvoiced sound waveform in the speech synthesizer according to the first embodiment of the present invention.

FIG. 17 is a diagram showing a state of expansion and contraction of a unit waveform spectrum in the speech synthesizer according to the second embodiment of the present invention.

FIG. 18 is a table showing an example of a scaling factor for each phoneme in the speech synthesizer according to the ninth embodiment of the present invention.

FIG. 19 is a table showing an example of changing the expansion / contraction ratio in the speech synthesizer according to Embodiment 9 of the present invention.

FIG. 20 is a block diagram showing a configuration of a conventional speech synthesizer.

21 is a diagram for explaining the operation of the speech synthesis device shown in FIG.

[Explanation of symbols]

 Reference Signs List 1 synthesis unit 1a syntax analysis unit 1b prosody giving unit 1c waveform editing unit 2 waveform conversion unit 2a parameter storage unit 2b waveform expansion / contraction unit 2c frequency conversion unit 2d frequency shift unit 2e frequency inverse conversion unit 3 waveform storage unit 4 input unit 5 second unit Storage unit 6 parameter extraction unit 7 conversion unit 7a parameter change unit 7b progress storage unit 7c waveform conversion unit

Claims (11)

[Claims] 1. A waveform storage unit for storing a unit waveform as a minimum unit of synthesis, which is a waveform for one pitch for voiced sounds and a waveform for one phoneme for unvoiced sounds. In a speech synthesizer having a waveform conversion unit that converts a stored unit waveform and a synthesis unit that edits the unit waveform converted by the waveform conversion unit to generate a synthesized sound, A speech synthesizer characterized in that the speech quality of the synthesized voice is changed independently of the pitch and the utterance speed by converting each of the voice segments individually and generating a synthesized voice using the converted voice waveform. 2. A waveform storage unit for storing a unit waveform as a minimum unit of synthesis, which is a waveform for one pitch for voiced sounds and a waveform for one phoneme for unvoiced sounds. In a speech synthesizer having a waveform conversion unit for converting a stored unit waveform and a synthesis unit for editing a unit waveform converted by the waveform conversion unit to generate a synthesized sound, the waveform conversion unit may include: A waveform expanding / contracting unit that expands / contracts a unit waveform with time, and a parameter storage unit that stores a parameter indicating a ratio of time expansion / contraction in the expanding / contracting unit, and synthesizes by time-expanding / dividing the unit waveform in the waveform expanding / contracting unit. A speech synthesizer characterized by changing the voice quality of sound. 3. A waveform storage unit for storing a unit waveform as a minimum unit of synthesis, which is a waveform for one pitch for voiced sounds and a waveform for one phoneme for unvoiced sounds. In a speech synthesizer having a waveform conversion unit for converting a stored unit waveform and a synthesis unit for editing a unit waveform converted by the waveform conversion unit to generate a synthesized sound, the waveform conversion unit may include: A frequency conversion unit that performs a Fourier transform on the unit waveform, a frequency shift unit that shifts the frequency of the unit waveform spectrum that has been Fourier-transformed by the frequency conversion unit, and a parameter necessary for performing a frequency shift by the frequency shift unit. A parameter storage unit, and a frequency inverse transform unit that performs Fourier inverse transform on the unit waveform spectrum frequency-shifted by the frequency shift unit. A speech synthesizer characterized by changing a voice quality of a synthesized sound by frequency-shifting a waveform spectrum. 4. A waveform storage unit for storing a unit waveform as a minimum unit of synthesis, which is a waveform for one pitch for voiced sounds and a waveform for one phoneme for unvoiced sounds. In a speech synthesizer having a waveform conversion unit for converting a stored unit waveform and a synthesis unit for editing a unit waveform converted by the waveform conversion unit to generate a synthesized sound, the waveform conversion unit may include: A frequency conversion unit that Laplace transforms a unit waveform, a frequency shift unit that shifts the frequency of a unit waveform spectrum that has been Laplace-transformed by the frequency conversion unit, and a parameter necessary for frequency shifting by the frequency shift unit. A parameter storage unit for storing, and a frequency inverse transform unit for performing an inverse Laplace transform of the unit waveform spectrum frequency-shifted by the frequency shift unit. A speech synthesizer characterized by changing a voice quality of a synthesized sound by frequency-shifting a half-wave spectrum. 5. A waveform storage unit for storing a unit waveform as a minimum unit of synthesis, which is a waveform for one pitch for voiced sounds and a waveform for one phoneme for unvoiced sounds. In a speech synthesizer having a waveform conversion unit for converting a stored unit waveform and a synthesis unit for editing a unit waveform converted by the waveform conversion unit to generate a synthesized sound, the waveform conversion unit may include: A matrix operation unit that integrates a matrix into the sample value column vector of the unit waveform stored in the waveform storage unit, and a matrix operation unit that integrates the matrix into the sample value column vector of the unit waveform in which the matrix is integrated, A speech synthesizer, comprising: a parameter storage unit that stores a matrix used in the matrix operation unit; and converting a unit waveform by integrating the matrix with a sample value sequence vector of the unit waveform. 6. The speech synthesizer according to claim 1, wherein the unit waveforms are divided into a plurality of groups,
A speech synthesizing device characterized in that conversion is performed using different parameters for each group. 7. The speech synthesizer according to claim 6, wherein the unit waveform is divided into a voiced sound and an unvoiced sound, and only the voiced sound is converted. 8. The speech synthesizing apparatus according to claim 1, further comprising a parameter input section, wherein the parameters can be set arbitrarily. 9. The speech synthesizer according to claim 1, wherein said waveform converter converts a unit waveform using a parameter specified in an input text. Synthesizer. 10. The speech synthesizer according to claim 1, further comprising: a second waveform storage unit that stores the unit waveform converted by the waveform conversion unit, wherein the second waveform storage unit stores the second waveform storage unit. A speech synthesizer for generating a synthesized speech using the unit waveform stored in the section. 11. The speech synthesizer according to claim 1, wherein the waveform converter changes a parameter from prosody information of the synthesizer.