JP3281266B2 - Speech synthesis method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech synthesis method for text speech synthesis, and more particularly to a speech synthesis method for generating a speech signal from information such as a phoneme symbol string, a pitch, and a phoneme duration.

[0002]

2. Description of the Related Art Creating a speech signal artificially from an arbitrary sentence is called text-to-speech synthesis. Text-to-speech synthesis is generally performed in three stages: a language processing unit, a phonemic processing unit, and a speech synthesis unit. The input text is first subjected to morphological analysis and syntactic analysis in the language processing unit, and then to accent and intonation processing in the phonological processing unit, resulting in phonological symbol strings, pitch, phonological duration, etc. Is output. Finally, the speech signal synthesizer synthesizes a speech signal from information such as a phoneme symbol string, pitch, and phoneme duration. Therefore, the speech synthesis method used for text speech synthesis must be a method capable of synthesizing an arbitrary phoneme symbol string with an arbitrary prosody.

[0003] The principle of such a speech synthesizer for synthesizing an arbitrary phoneme symbol string is that if a vowel is represented by V and a consonant is represented by C, a characteristic parameter of a small basic unit such as CV, CVC, VCV (which is referred to as After representative voice units are stored and selectively read out, the voices are synthesized by controlling the pitch and the duration to connect. Therefore, the stored representative speech unit greatly affects the quality of the synthesized speech.

Heretofore, the production of these representative speech units has relied solely on humans, and most of them are cut out from the speech signal by trial and error, which required a great deal of labor. As a method of automating such a representative speech unit creation operation and easily generating a representative speech unit suitable for use in speech synthesis, for example, a technique called phoneme environment clustering (COC) is disclosed in −7830
0 "Speech synthesis method".

[0005] The principle of the COC is that a phoneme name or a label of a phoneme environment is assigned to a large number of speech units, and a plurality of speech units to which the label is assigned are assigned to a plurality of speech units based on a distance measure between the speech units. The speech is classified into clusters, and the centroid of each cluster is used as a representative speech unit. Here, the phoneme environment is a combination of all factors that are an environment for the speech unit, and the factors include a phoneme name, a preceding phoneme, a succeeding phoneme, a subsequent phoneme, a pitch period, power, and the like of the speech unit. The presence or absence of stress, the position from the accent nucleus, the time since breathing, the utterance speed, the emotion, and the like can be considered. Each phoneme in the real speech changes its phoneme depending on the phoneme environment. By storing representative speech units for each of multiple clusters related to the phoneme environment, natural speech taking into account the influence of the phoneme environment is synthesized. It is possible to do.

[0006]

As described above, in speech synthesis for text speech synthesis, it is necessary to change the pitch and duration of representative speech units to specified values for synthesis. . Due to such a change in the pitch and the duration, the sound quality of the synthesized speech is degraded to some extent as compared with the sound quality of the speech signal from which the representative speech unit has been cut out.

On the other hand, in the above-described clustering based on COC, since the clustering is merely performed based on the distance scale between speech units, the effect of changing the pitch and duration during synthesis is completely considered. There is a problem that is not. That is, the clustering by the COC and the representative speech unit of each cluster are actually performed at the level of the synthesized speech synthesized by changing the pitch and the duration.
There is no guarantee that it will be appropriate.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a speech synthesis method capable of effectively improving the sound quality of a synthesized speech by text speech synthesis. .

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention considers the influence of a change in pitch or duration, and reduces the representative voice to reduce distortion of natural voice at the level of synthesized voice. By generating a segment and synthesizing a speech using the representative speech segment, a synthesized speech close to a natural speech is generated.

That is, according to the present invention, at least one of the pitch and the duration of each of a plurality of input speech units (second speech units) is set to a plurality of training speech units which are the same as or different from the input speech units. (First speech unit)
A plurality of synthesized speech units are generated by changing them to be equal to at least one of the pitch and the duration length of any of the above, and between each of the synthesized speech units and each of the training speech units. The distortion of each of the synthesized speech units is evaluated based on a distance measure, and a set of a predetermined number of speech units is represented from the input speech unit based on a distortion evaluation function using the distortion information. A synthesized speech is generated by selecting and storing as a unit, selecting and connecting a representative unit from the stored representative unit according to an input phoneme.

Here, the first and second speech units are C
It is a segment cut out from a speech signal in a speech synthesis unit such as V, VCV, or CVC, and represents a segmented waveform or a parameter series extracted from the waveform by some method. Among these, the first speech unit is used to evaluate the distortion of the synthesized speech, and the second speech unit is used as a representative speech unit candidate.
The synthesized speech unit represents a synthesized speech waveform or a parameter sequence generated by changing at least the pitch or the duration of the second speech unit.

The distance measure between the synthesized speech unit and the first speech unit represents the distortion of the synthesized speech. Therefore, a speech unit having a smaller distance scale, that is, a distortion, is selected from the second speech units and stored as representative speech units, and a predetermined representative speech unit is selected from these representative speech units. Connection, a high-quality synthesized speech close to natural speech is generated.

In a first aspect of the present invention, the pitch and duration of a plurality of second speech units are determined according to at least one of the pitch and duration of the plurality of first speech units labeled with a phoneme environment. A plurality of synthesized speech units are generated by changing at least one of the lengths, and a plurality of representative speeches are generated from the second speech unit based on a distance measure between the synthesized speech units and the first speech unit. A speech unit is selected and stored, a plurality of phoneme environment clusters corresponding to the representative speech units are respectively generated based on the distance scale, and a phoneme environment cluster including a phoneme environment of an input phoneme from the plurality of representative speech units. Is synthesized by selecting and connecting a representative speech unit corresponding to.

Here, the phoneme environment is a factor which is an environment for a speech unit as described above, such as a phoneme name of the speech unit, a preceding phoneme, a succeeding phoneme, a succeeding phoneme, a pitch period,
It is a combination of elements such as power, the presence or absence of stress, the position from the accent nucleus, the time since breathing, the utterance speed, and the emotion.In other words, the phoneme environment cluster is a set of phoneme environments. / Ka /,
Preceding phoneme is / i / or / u /, pitch frequency is 200
以下 z or less ”.

As in the first embodiment, a plurality of phoneme environment clusters corresponding to the representative speech units are generated based on the distance scale, ie, the distortion of the synthesized speech, and the phoneme environment clusters including the phoneme environment of the input phoneme are generated. If a representative speech unit is selected and connected, for example, even if a speech unit with the same phoneme name exists in multiple phoneme environments, it corresponds to a phoneme environment cluster that includes the phoneme environment of the actual input phoneme By selecting only the representative speech unit to be synthesized, a more natural synthesized speech can be obtained.

According to a second aspect of the present invention, the pitch and duration of the plurality of second speech units are determined according to at least one of the pitch and duration of the plurality of first speech units labeled with a phoneme environment. Generating a plurality of synthesized speech units by changing at least one of the lengths; generating a plurality of phoneme environment clusters based on a distance measure between the synthesized speech units and the first speech unit; A plurality of representative speech units respectively corresponding to each phoneme environment cluster are selected and stored from the second speech unit based on the distance scale, and a predetermined representative speech unit is selected from these representative speech units. Synthesize voice by connecting. The second aspect is effective when a speech unit exists only in one phoneme environment.

According to a third aspect of the present invention, the pitch and duration of a plurality of second speech units are determined according to at least one of the pitch and duration of the plurality of first speech units labeled with a phoneme environment. Generating a plurality of synthesized speech units by changing at least one of the lengths; generating a plurality of phoneme environment clusters based on a distance measure between the synthesized speech units and the first speech unit; A plurality of representative speech units corresponding to each of the phoneme environment clusters are selected and stored from the second speech unit based on a distance scale between the one speech unit and the synthesized speech unit, and these representative speech units are stored. A speech is synthesized by selecting and connecting a representative speech unit corresponding to a phoneme environment cluster including a phoneme environment of an input phoneme from speech units.

According to the third embodiment, similarly to the first embodiment, for example, when a speech unit having the same phoneme name exists in a plurality of phoneme environments, a phoneme environment cluster including a phoneme environment of an actual input phoneme is included. By selecting only the representative speech unit corresponding to, a more natural synthesized speech can be obtained.

Another speech synthesizing method according to the present invention comprises:
Generating a plurality of synthesized speech units by changing at least one of the pitch and the duration of the plurality of second speech units according to at least one of the pitch and the duration of the plurality of first speech units; Further, these synthesized speech units are subjected to spectrum shaping, and a plurality of synthesized speech units are subjected to spectrum shaping based on a distance scale between each synthesized speech unit and the first speech unit. The selected speech unit is selected and stored, and a predetermined representative speech unit is selected from these representative speech units and connected to synthesize a speech. It is characterized by generating a synthesized speech.

In this case, also in the first, second and third modes described above, after a plurality of synthesized speech units are generated, spectrum shaping is performed. Here, the spectral shaping is a process for synthesizing a clear voice with a “slippery”, and is realized by, for example, filtering by an adaptive post filter that performs formant emphasis and pitch emphasis.

In this manner, the spectrum synthesized for the speech synthesized by the connection of the representative speech units is performed, and the same spectrum shaping is performed for the synthesized speech unit, so that the final after the spectrum shaping is performed. At the level of the synthesized speech, a representative speech segment that can reduce distortion with respect to natural speech can be generated, so that a clearer synthesized speech excellent in “turning” can be obtained.

In the present invention, the representative speech unit is
Information of a sound source signal and a set of coefficients of a synthesis filter that generates a synthesized speech signal by using the sound source signal as an input may be stored. In this case, if the sound source signal and the coefficients of the synthesis filter are quantized, and information of a set of the quantized sound source signal and the coefficient of the synthesis filter is stored, the sound source signal stored as the representative speech unit and the synthesis filter Since the number of coefficients can be reduced, the calculation time required for learning a synthesis unit is shortened, and the amount of memory required for actual speech synthesis is reduced.

Further, the number of at least one of the sound source signal and the coefficient of the synthesis filter stored as information on the representative speech unit can be smaller than the total number of speech synthesis units or the total number of phoneme environment clusters. Even in this way, a good synthesized speech can be obtained.

[0024]

Embodiments 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 speech synthesis apparatus for realizing a speech synthesis method according to a first embodiment of the present invention. This speech synthesizer is roughly divided into a synthesis unit learning system 1 and a rule synthesis system 2. When text text synthesis is actually performed, the rule synthesis system 2 operates, and the synthesis unit learning system 1 performs learning in advance to generate a representative speech unit.

First, the synthesis unit learning system 1 will be described. The synthesis unit learning system 1 includes a representative speech unit generation unit 11 that generates a representative speech unit and a phoneme environment cluster associated therewith.
And a representative speech unit storage unit 12 and a phoneme environment cluster storage unit 13. The representative speech unit generation unit 11 includes a training speech unit 101 that is a first speech unit.
And a phoneme environment 102 and an input speech unit 103 as a second speech unit.

In the representative speech unit generation unit 11, according to the information of the pitch period and the duration length included in the phoneme environment 102 labeled on the training speech unit 101,
By changing the pitch period and duration of the input speech unit 103, a plurality of synthesized speech units are internally generated,
Further, these synthesized speech unit and training speech unit 1
A representative speech unit 104 and a phoneme environment cluster 105 are generated according to a distance scale of 01. Phoneme environment cluster 1
05 is generated by classifying the training speech units 101 into clusters relating to phoneme environments, as described later.

The representative speech unit 104 is stored in the representative speech unit storage unit 12, and the phoneme environment cluster 105 is associated with the representative speech unit 104 and stored in the phoneme environment cluster storage unit 13.
Is stored. The process of the representative speech unit generation unit 11 will be described later in detail.

Next, the rule synthesizing system 2 will be described. The rule synthesis system 2 includes a representative speech unit storage unit 12, a phoneme environment cluster storage unit 13, a unit selection unit 14, and a speech synthesis unit 15.
, And shares the representative speech unit storage unit 12 and the phoneme environment cluster storage unit 13 with the synthesis unit learning system 1.

The segment selection unit 14 includes, as input phoneme information, prosody information 111 and phonological symbols obtained by morphological analysis / syntax analysis of the input text for text-to-speech synthesis and further performing accent and intonation processing. Column 112 is entered. The prosody information 111 includes a pitch pattern and a phoneme duration. The unit selection unit 14 stores the prosody information 111 and the phoneme symbol string 112
Internally generate the phoneme environment of the input phoneme from

Then, the segment selecting unit 14 reads the phoneme environment cluster 106 read from the phoneme environment cluster storage unit 13.
, The phoneme environment of the input phoneme belongs to which phoneme environment cluster, and the representative speech unit selection information 107 corresponding to the searched phoneme environment cluster is output to the representative speech unit storage unit 12.

The speech synthesizer 15 applies the prosody information 11 to the representative speech unit 108 selectively read from the representative speech unit storage unit 12 in accordance with the representative speech unit selection information 107.
In accordance with 1, the pitch period and the phoneme duration are changed, and segments are connected to output the synthesized speech signal 113. Here, as a method of synthesizing voice by connecting the segments by changing the pitch and the duration time, a known technique such as a residual driving LSP method or a waveform editing method can be used.

Next, the processing procedure of the representative speech unit generation unit 11 which is a feature of the present invention will be specifically described. The flowchart of FIG. 2 shows a first processing procedure of the representative speech unit generation unit 11.

In the representative speech unit generation processing according to the first embodiment, a large number of continuously uttered speech data are labeled for each phoneme as a preparatory stage, and CV,
A training speech unit T _i (i = 1, 2, 3,..., N _T ) is cut out according to a synthesis unit such as VCV and CVC. Also, a phoneme environment P _i (i = 1, 2, 3,..., _NT ) corresponding to each training speech unit T _i is extracted. Here, _NT represents the number of training speech segments. The phoneme environment P _i shall include at least information on the phoneme of the training speech unit T _i , its pitch and duration, and, if necessary, information on the preceding and following phonemes.

Next, the above described training speech unit T _i
Of the input speech units S _j
(I = 1, 2, 3,..., N _S ). Where N
_S represents the number of input speech units. Here, the same input speech unit S _j as the training speech unit T _i may be used (that is, T _i = S _i ), or a speech unit different from the training speech unit T _i may be created. May be. In any case, it is desirable that a large number of training speech units and input speech units having a rich phonemic environment are prepared.

After passing through such a preparation stage, first, in a speech synthesis step S21, the input speech unit S _j is set to be equal to the pitch and duration included in the phoneme environment P _i.
The synthesized speech unit G _ij is generated by synthesizing speech by changing the pitch and the duration of the speech. Here, the change of the pitch and the duration is performed in the same manner as the change of the pitch and the duration in the voice synthesizer 15. All phoneme environments P _i (i = 1, 2, 3,
, N _T ), the input speech unit S _i (j = 1, 2, 2)
3,..., N _S )
N _T × N _S synthesized speech units G _ij (i = 1, 2, 3,
_{..., N T, j = 1,2,3} , ..., to generate the N _S).

Next, the distortion evaluation step S22, the evaluation of the strain e _ij synthetic speech unit G _ij. The evaluation of the distortion e _ij is based on the synthesized speech unit G _ij and the training speech unit T _i.
This is done by finding a distance measure between. Any spectral distance can be used for the distance measure. For example, a synthesized speech unit G _ij and a training speech unit T
_{For i} , a method of obtaining a power spectrum using FFT (Fast Fourier Transform) or the like to evaluate the distance between each power spectrum, or performing a linear prediction analysis to obtain LP
There is a method of obtaining the C or LSP parameters and evaluating the distance between the parameters. In addition, a method of evaluating using a transform coefficient such as a short-time Fourier transform or a wavelet transform can be used. Alternatively, a method of evaluating the distortion after normalizing the power of each segment may be used.

Next, in a representative speech unit generation step S23, based on the distortion e _ij obtained in step S22, the representative speech units of the number 代表 of the representative speech units designated from the input speech units _Sj. A piece D _k (k = 1, 2, 3,..., N) is selected.

An example of the representative speech unit selection method will be described. Set of N speech units are selected from the input speech segments _{S j U = {u k |} u k = S j (k = 1,2,3, ...,
N) 評 価, an evaluation function E _D1 (U) representing the sum of distortions
Is defined as in the following equation (1).

[0039]

(Equation 1)

Where min (e _ij1 , e _ij2 , e
_ij3 , ..., _eijN ) are _eij1 , _eij2 , _eij3 , ..., e
_This is a function representing the minimum value in _ijN . The combination of the set U is N _S! / @ N! (N _S -N)! There are｝ ways, and a set U of these speech units is searched for a U that minimizes the evaluation function E _D1 (U), and its element u _k is set as a representative speech unit D _k .

Finally, a phoneme environment cluster generation step S
At 24, a plurality of clusters (phoneme environment clusters) C _k (k = 1, 2, 3,...) Related to the phoneme environment are generated from the phoneme environment P _i , the distortion e _ij, and the representative speech unit D _k . The phoneme environment cluster C _k is obtained, for example, by searching for a cluster that minimizes the clustering evaluation function E _C1 represented by the following equation (2).

[0042]

(Equation 2)

The representative speech unit D _k and the phoneme environment cluster C _k generated in steps S23 and S24 in this way.
Are stored in the representative speech unit storage unit 12 and the phoneme environment cluster storage unit 13 in FIG.

Next, the second processing procedure of the representative speech unit generation unit 11 will be described with reference to the flowchart of FIG. In the representative speech unit generation process according to the second processing procedure, first, in an initial phoneme environment cluster generation step S30, clustering of phoneme environments is performed in advance based on some foresight knowledge to generate an initial phoneme environment cluster. For the clustering of the phoneme environment, for example, clustering based on phonemes can be performed.

Then, of the input speech unit S _j and the training speech unit T _i , only the speech units having the same phoneme are used, and steps S21, S22, S2 in FIG.
3, a synthetic speech unit generation step S31 similar to S24,
Distortion evaluation step S32, representative speech unit generation step S
33, the processing of the phoneme environment cluster generation step S34 is sequentially performed, and the same operation is repeated for all the initial phoneme environment clusters, thereby generating all the representative speech units and the corresponding phoneme environment clusters. The representative speech unit and the phoneme environment cluster thus generated are stored in the representative speech unit storage unit 12 and the phoneme environment cluster storage unit 13 in FIG. 1, respectively.

However, if the number of representative speech units per each initial phoneme environment cluster is 1, the initial phoneme environment cluster becomes the phoneme environment cluster of the representative speech unit, so that the phoneme environment cluster generation step S34 becomes unnecessary. The initial phoneme environment cluster may be stored in the phoneme environment cluster storage unit 13.

Next, the third processing procedure of the representative speech unit generation unit 11 will be described with reference to the flowchart of FIG. In the representative speech segment generation processing according to the third processing procedure, the speech synthesis step S is performed in the same manner as in the first processing procedure shown in FIG.
41 and a distortion evaluation step S42 in order, in a next phoneme environment cluster generation step S43, a cluster C _k (k = 1, 2, 3,..., に 関 す る) relating to the phoneme environment based on the phoneme environment P _i and the distortion e _ij. ). The phoneme environment cluster C _k is obtained, for example, by searching for a cluster that minimizes the clustering evaluation function E _C2 represented by the following equations (3) and (4).

[0048]

(Equation 3)

Next, representative in speech unit generation step S44, the distortion corresponding to each phoneme environment cluster C _k based on e _ij representative speech units D _k input speech segment S _j
Choose more. This representative speech unit D _k is _obtained from the input speech unit S _j by, for example, a distortion evaluation function E _D2 represented by the following equation (5).
(j) is obtained by searching for a speech unit that minimizes.

[0050]

(Equation 4)

The representative speech unit generation process according to the third processing procedure is modified so that, similar to the second processing procedure, the representative speech unit cluster is generated for each initial phoneme environment cluster generated in advance based on some foresight knowledge. It is also possible to generate segments and generate phoneme environment clusters.

(Second Embodiment) Next, a second embodiment of the present invention will be described with reference to FIGS. FIG.
FIG. 3 is a block diagram illustrating a configuration of a speech synthesis device that realizes a speech synthesis method according to a second embodiment. 1 will be described with the same reference numerals attached thereto. The first embodiment differs from the first embodiment in that an adaptive post filter 16 is added after the speech synthesizer 15 in the present embodiment. Unlike the embodiment, in addition to this, the method of generating a plurality of synthesized speech units in the representative speech unit generation unit 11 is also different from the previous embodiment.

That is, as in the first embodiment, the representative speech unit generation unit 11 inputs the input speech in accordance with the pitch period and duration information included in the phoneme environment 102 labeled on the training speech unit 101. Element 103
After a plurality of synthesized speech units are internally generated by changing the pitch period and the duration time, the synthesized speech units are subjected to filtering by an adaptive post-filter to perform spectrum shaping. Then, a representative speech unit 104 and a phoneme environment cluster 105 are generated in accordance with a distance scale between each synthesized speech unit and the training speech unit 101 after spectrum shaping by the adaptive post filter. The phoneme environment cluster 105 is generated by classifying the training speech units 101 into clusters relating to the phoneme environment, as in the previous embodiment.

The representative speech unit generation unit 11 changes the pitch period and the duration of the input speech unit 103 in accordance with the information on the pitch period and the duration included in the phoneme environment 102. The adaptive post-filter that performs filtering on the synthesized speech unit to perform spectrum shaping may have the same configuration as the adaptive post-filter 16 disposed downstream of the speech synthesis unit 15.

On the other hand, the speech synthesizing unit 15 applies a prosody to the representative speech unit 108 selectively read from the representative speech unit storage unit 12 according to the representative speech unit selection information 107 as in the first embodiment. The synthesized speech signal 113 is generated by changing the pitch period and the phoneme duration according to the information 111 and connecting the segments. In the present embodiment, the synthesized speech signal 113 is further processed by the adaptive post-filter 1.
6, where the spectrum is shaped to improve the sound quality, and the final synthesized speech signal 114 is extracted.

FIG. 6 shows an example of the configuration of the adaptive post filter 16. The adaptive post filter 16 is configured by arranging a formant emphasis filter 21 and a pitch emphasis filter 22 in cascade.

The formant emphasis filter 21 is determined based on the LPC coefficient obtained by performing the LPC analysis on the representative speech unit 108 selectively read from the representative speech unit storage unit 12 according to the representative speech unit selection information 107. According to the filter coefficient to be applied, the synthesized speech signal 113 input from the speech synthesis unit 15 is filtered to enhance the spectrum peak. On the other hand, the pitch emphasis filter 22 performs a process of emphasizing the pitch of the audio signal by filtering the output of the formant emphasis filter 21 according to a parameter determined based on the pitch cycle included in the prosody information 111. In addition,
Formant emphasis filter 21 and pitch emphasis filter 22
May be reversed.

The spectrum is shaped by the application of the adaptive post-filter 16, and a synthesized voice signal 114 capable of reproducing a clear voice with a sharp edge is obtained. The adaptive postfilter 16 is not limited to the configuration shown in FIG. 6, but may employ various configurations based on known techniques used in the fields of voice coding and voice synthesis.

As described above, in this embodiment, the rule synthesizing system 2
At the subsequent stage of the speech synthesizing unit 15,
In consideration of the fact that 6 is arranged, in the synthesis unit learning system 1 as well, the pitch period of the input speech unit 103 and the pitch period of the input speech unit A plurality of synthesized speech units generated by changing the duration are similarly filtered by an adaptive post filter. Therefore,
At the same level as the final synthesized speech signal 114 after passing through the adaptive post filter 16, a representative speech unit that reduces distortion with respect to natural speech is represented by the representative speech unit generation unit 1.
1, it is possible to generate a synthesized voice closer to a natural voice.

Next, the representative speech unit generator 1 in FIG.
The first processing procedure will be specifically described. The flowcharts of FIGS. 7, 8 and 9 show the first, second and third processing procedures of the representative speech unit generation unit 11 in FIG. 7, 8, and 9, post-filtering steps S25, S36, and S45 are added after the speech synthesis steps S21, S31, and S41 in the processing procedure shown in FIGS. 2, 3, and 4 described above. ing.

Post-filtering steps S25, S
In 36 and S45, the filtering by the above-mentioned adaptive post filter is performed. That is, for speech synthesis step S21, S31 and S41 synthesized speech segment G _ij generated in performs filtering according to filter coefficients determined based on the input speech segments S _i to the LPC coefficients obtained by LPC analysis In this way, formant emphasis is performed to emphasize the peaks of the spectrum. Further, the synthesized speech unit after the formant emphasis is further filtered in accordance with a parameter determined based on the pitch cycle of the training speech unit T _i , thereby performing pitch emphasis.

In this way, spectrum shaping is performed in the post-filtering steps S25, S36 and S45. This post-filtering step S2
5, S36 and S45 are to perform post-filtering for improving the sound quality by shaping the spectrum of the synthesized voice signal 113 by the adaptive post-filter 16 provided in the subsequent stage of the voice synthesis unit 15 in the rule synthesis system 2 as described above. Is a process that enables learning of a synthesis unit. By combining this process with the process by the adaptive post-filter 16, a clear synthesized speech signal 114 having a "finish" is finally generated. .

(Third Embodiment) Next, FIGS.
The third embodiment of the present invention will be described with reference to FIG. FIG.
FIG. 9 is a block diagram illustrating a configuration of a synthesis unit learning system in the speech synthesis device according to the second embodiment.

The synthesis unit learning system 30 in the present embodiment
Are the LPC analysis / inverse filter unit 31 and the sound source signal storage unit 3
2. LPC coefficient storage unit 33, sound source signal generation unit 34, synthesis filter 35, distortion calculation unit 36, and minimum distortion search unit 17
Consists of zero. A training speech unit 101, a phoneme environment 102 and an input speech unit 103 labeled with the training speech unit 101 are input to the synthesis unit learning unit 30. The input speech unit 103 is input to the LPC analysis / inverse filter unit 31, where LPC analysis is performed, and an LPC coefficient 201 and a prediction residual signal 202 are output. LPC coefficient 201 is LPC
The prediction residual signal 202 is stored in the coefficient storage unit 33, and the prediction residual signal 202 is stored in the excitation signal storage unit 32.

The prediction residual signals stored in the sound source signal storage unit 32 are read out one by one in accordance with a command from the minimum distortion search unit 37, and are read by the sound source signal generation unit 34 into the phoneme environment 102 of the training speech unit 101. According to the information on the pitch pattern and the phoneme duration included, the pitch period and the duration are changed to generate a sound source signal. The sound source signal generated in this manner is input to a synthesis filter 35 using the LPC coefficient read from the LPC coefficient storage unit 33 as a filter coefficient in accordance with an instruction from the minimum distortion search unit 37 to generate a synthesized speech unit. Is done.

Next, an error, that is, a distortion of the synthesized speech unit with respect to the training speech unit 101 is calculated in the distortion calculating unit 36, and this distortion is evaluated in the minimum distortion searching unit 37. The minimum distortion search unit 37 issues a command to output all combinations of LPC coefficients and prediction residual signals stored in the LPC coefficient storage unit 33 and the excitation signal storage unit 32, respectively, and responds to those combinations. Then, the synthesis speech unit is generated by the synthesis filter 35. Then, a combination of the LPC coefficient that gives the minimum distortion and the prediction residual signal is found and stored.

Next, the operation of the synthesis unit learning system 30 will be described with reference to the flowchart of FIG. First, as a preparation stage, labeling is performed on a large number of continuously uttered speech data for each phoneme, and a training speech unit T _i (i = 1,2,2,3) is obtained in accordance with a synthesis unit such as CV, VCV, or CVC.
3,..., N _T ). Further, a phoneme environment P _i (i = 1, 2, 3, 3) corresponding to each training speech unit T _i
.., _NT ) are extracted in advance. However, N _T is the number of training speech units T _i. The phoneme environment includes at least the phoneme of the training speech unit, its pitch pattern and duration, and also includes the preceding and following phonemes as necessary.

Next, a large number of input speech units S _i (i =
1, 2, 3,..., N _S ). However, N _S represents the number of input speech segments S _i. Here, the input speech unit S
_Assume that the synthesis unit of _i and the training speech unit T _i match. For example, when creating a representative speech unit of a certain CV syllable “ka”, an input speech unit S _i and a training speech unit T _i are set from syllables “ka” cut out from a large number of speech data. The input speech unit may be the same as the training speech unit (that is, T _i = S _i ), or a different speech unit may be created.
In any case, it is desirable that a large number of training speech units and input speech units having a rich phonemic environment be prepared.

After such a preparation stage, the input speech unit S _i (i = 1, 2, 3,...,
N _S ), and an LPC coefficient a _i (i = 1,
2, 3,..., N _s ) and perform inverse filtering based on the coefficients to obtain a prediction residual signal e _i (i =
1, 2, 3,..., N _S ). Where a is p
As an order of PC analysis, it is a vector having p elements.

Next, the obtained prediction residual signal is stored as a sound source signal in step S52 together with the LPC coefficient. Next LPC coefficient / sound source signal combination step S5
In step 3, a set (a _i , e _j ) of a combination of the stored LPC coefficient and the sound source signal is created.

[0071] for the combination of the set, to generate a sound source signal by modifying the pitch and duration of e _j to be equal to the pitch pattern and duration of P _k in the next speech synthesis step S54 Thereafter, a filtering operation is performed by a synthesis filter having LPC coefficients a _i to generate a synthesized speech unit G _k (i, j).

As described above, all P _k (k = 1, 2, 2)
3,..., N _T ) by performing speech synthesis.
_T synthesized speech units G _k (i, j), (k = 1, 2,
3,..., N _T ).

In the next distortion evaluation step S55, the sum E of the distortion E _k (i, j) between the synthesized speech unit G _k (i, j) and the training speech unit T _k and the distortion E related to P _k is calculated. It is determined by the following equations (6) and (7).

[0074]

(Equation 5)

Here, D is a distortion function, and any spectral distance can be used. For example, there is a method of obtaining a power spectrum by using FFT or the like and obtaining a distance therebetween, or a method of obtaining a LPC or LSP parameter by performing linear prediction analysis and evaluating a distance between parameters. In addition, there is a method of evaluating using a transform coefficient such as a short-time Fourier transform or a wavelet transform. It is also conceivable to evaluate the distortion after normalizing the power of each element.

The processing of steps S53 to S55 is repeated for all combinations (a _i , e _j ) of LPC coefficients and sound source signals,
(I, j = 1, 2, 3,..., N _S ), and in a distortion evaluation step S55, a set of i, j that gives the minimum value of E (i, j) is searched.

In the next representative speech unit generation step S57, a set of i, j giving the minimum value of E (i, j), or the corresponding (ai, ej) or (ai, ej)
Is saved as a representative speech unit. However, this representative speech unit generation step is a process for generating one set of representative speech units for each synthesis unit. If N sets are to be generated, the following is performed. First, N _S * N _S
A set selected from N sets of (ai, ej) sets is set as equation (8), and an evaluation function representing the sum of distortions is defined as equation (9).

[0078]

(Equation 6)

Here, min () is a function representing the minimum value. There are N _S * N _S C _N combinations of the set U,
Evaluation function ED a (U) to explore the U to minimize from these set U, and that element (a _i, e _j) represent speech units to ^k.

Next, the rule synthesizing system in this embodiment will be described with reference to FIG. The rule synthesis system 40 according to the present embodiment includes a combination storage unit 41, a sound source signal storage unit 4
2. It comprises an LPC coefficient storage unit 43, a sound source signal generation unit 44, and a synthesis filter 45. The rule synthesis unit 40 receives prosody information 111 and a phoneme symbol string 112 obtained as a result of language processing of the input text and subsequent phoneme processing. Combination storage unit 41, sound source signal storage unit 42 and L
The PC coefficient storage unit 43 stores the synthesis unit learning unit 30 of FIG.
The combination information (i, j) of the LPC coefficient and the sound source signal obtained in the above, and the sound source signal e _j and the LPC coefficient a _i are stored in advance.

The combination storage unit 41 stores the phoneme symbol string 112
Is input and the corresponding synthesis unit (for example, CV syllable)
Is output as the combination information of the LPC coefficient and the excitation signal that gives The sound source signal stored in the sound source signal storage unit 42 is read out according to a command from the combination storage unit 41, and is read according to the pitch pattern and phoneme duration information included in the prosody information 111 input in the sound source signal generation unit 44. , Its pitch period and duration are changed, and connection of the sound source signal is performed.

The sound source signal thus generated is subjected to a synthesis filter 45 using the coefficient read from LPC coefficient storage section 43 as a filter coefficient in accordance with a command from combination storage section 41.
, And the interpolation of the filter coefficients and the filtering operation are performed, so that the synthesized speech signal 113 is created.

(Fourth Embodiment) Next, a fourth embodiment of the present invention will be described with reference to FIGS. FIG.
3 is a diagram showing a schematic configuration of a synthesis unit learning system according to the present embodiment, which has a configuration in which a clustering unit 38 is added to the synthesis unit learning system 30 shown in FIG. 10 of the third embodiment. This embodiment is different from the third embodiment in that the clustering unit 38 performs clustering of phoneme environments in advance based on some foresight knowledge and generates a representative speech unit for each cluster. As the clustering, for example, clustering based on the pitch of the segment can be considered. In this case, training speech unit 1
01 is clustered based on the pitch, and the representative speech unit described in the third embodiment is generated for the training speech unit of each cluster.

FIG. 14 is a diagram showing a schematic configuration of a rule synthesizing system according to the present embodiment, in which a clustering unit 48 is added to the rule synthesizing system 40 shown in FIG. 12 of the third embodiment. . The prosody information 111 is pitch-clustered in the same manner as the training speech segments, and speech is synthesized using the sound source signals and LPC coefficients corresponding to the representative speech segments of each cluster obtained by the synthesis unit learning system 30.

(Fifth Embodiment) Next, FIGS.
A fifth embodiment of the present invention will be described with reference to FIG. FIG.
Is a block diagram illustrating a synthesis unit learning system according to the present embodiment, and illustrates a configuration example in which a cluster is automatically generated based on a distortion measure with respect to a training speech unit. In the present embodiment, the synthesis unit learning system 30 shown in FIG.
In this configuration, a phoneme environment cluster generation unit 51 and a cluster storage unit 52 are added.

The first example of the synthesis unit learning system in this embodiment
Will be described with reference to the flowchart shown in FIG. This processing procedure is similar to the processing procedure of the third embodiment shown in FIG.
Has been added. In this step S58, a cluster C _m (m = 1, 2, 3, 3) relating to the phoneme environment is obtained from the phoneme environment P _k , the distortion E _k (i, j) and the representative speech unit D _m .
.., N). The phoneme environment cluster C _m is, for example, a clustering evaluation function E _cm represented by the following equation (10).
By searching for a cluster that minimizes

[0087]

(Equation 7)

FIG. 17 shows a second example of the synthesis unit learning system shown in FIG.
6 is a flowchart showing the processing procedure of FIG. In this processing, in the initial phoneme environment cluster generation step S50, the phoneme environments are clustered in advance based on some foresight knowledge to generate an initial phoneme environment cluster. As the clustering, for example, clustering based on phonemes of the segment can be considered. In this case, the generation of the representative speech unit and the generation of the phoneme environment cluster described in the third embodiment are performed using only the speech unit and the training speech unit in which the phonemes of the segment match. By repeating the same operation for environment clusters, all representative speech units and corresponding phoneme environment clusters are generated.

However, if the number of representative speech units per each initial cluster is 1, the initial phoneme environment cluster becomes the phoneme environment cluster of the representative speech unit, so that the phoneme environment cluster generation step S58 becomes unnecessary, and the initial phoneme The environment cluster may be stored in the cluster storage unit 52 of FIG.

The rule synthesizing system according to the present embodiment is the same as that shown in FIG.
Is configured similarly to the rule total system 40 in the fourth embodiment shown in FIG. In this case, the clustering unit 48
Processing is performed based on the information stored in the fifth cluster storage unit 52.

(Sixth Embodiment) FIG. 18 shows a configuration of a synthesis unit learning system according to a sixth embodiment of the present invention. The synthesis unit learning system according to the present embodiment has a configuration in which buffers 61 and 62 and quantization table creation units 63 and 64 are added to the synthesis unit learning system 30 shown in FIG.

In this embodiment, the input speech unit 103
Is input to the LPC analysis / inverse filter unit 31, where L
After the LPC coefficient 201 and the prediction residual signal 202 generated by the PC analysis are temporarily stored in the buffers 61 and 62, respectively, they are quantized by the quantization table creation units 63 and 64, respectively. The residual signal is L
They are stored in the PC coefficient storage unit 33 and the sound source signal storage unit 34, respectively.

FIG. 19 is a flowchart showing the processing procedure of the synthesis unit learning system of FIG. 18. The difference from the processing procedure shown in the flowchart of FIG. 11 is that a quantization step S60 is added after the LPC analysis step S51. That is. In this quantization step S60, the LPC coefficient a _i (i =
, N _s ) and the prediction residual signal e _i (1 = 1,
2, 3,..., N _S ) are temporarily stored in the buffer, and then LB
A quantization table is created using a known technique such as the G algorithm, and the LPC coefficient and the prediction residual signal are quantized.
At this time, the size of the quantization table, i.e. the number of the representative vector quantization is less than N _S. Then, the quantized LPC coefficient and the prediction residual signal are stored in the next step S52. Subsequent processing is the same as in the case of FIG.

(Seventh Embodiment) FIG. 20 is a block diagram showing a synthesis unit learning system according to a seventh embodiment of the present invention, in which clusters are automatically generated based on a distortion measure with respect to a training speech unit. 4 shows a configuration example in the case of generation. Cluster generation can be performed in the same manner as in the fifth embodiment. That is, the synthesis unit learning system according to the present embodiment has a configuration in which the fifth embodiment illustrated in FIG. 15 and the sixth embodiment illustrated in FIG. 18 are combined.

(Eighth Embodiment) FIG. 21 shows a synthesis unit learning system according to an eighth embodiment of the present invention, in which an LPC analysis section 31a and an inverse filter 31b are separated and a buffer 6
1 shows an example of a configuration in which inverse filtering is performed using the LPC coefficients quantized via the quantization table creation unit 63 and the prediction residual signal is calculated. By doing so, it is possible to generate a representative speech unit that reduces sound quality degradation of synthesized speech due to quantization distortion of LPC coefficients.

(Ninth Embodiment) FIG. 22 shows a synthesis unit learning system according to a ninth embodiment of the present invention. As in the eighth embodiment, inverse filtering is performed using quantized LPC coefficients. 14 shows another configuration example in the case of performing ringing and calculating a prediction residual signal. However, the eighth embodiment is different from the eighth embodiment in that the prediction residual signal inversely filtered by the inverse filter 31b is quantized via the buffer 62 and the quantization table 64 and then input to the excitation signal storage unit 32. Is different.

In the sixth to ninth embodiments, the size of the quantization table created by the quantization table creation units 63 and 64, that is, the number of representative spectra for quantization is determined by the number of clusters or the total number of synthesis units. (For example, the total number of CV and VC syllables). Thus LP
By quantizing the C coefficient and the prediction residual signal, it is possible to reduce the number of LPC coefficients and the number of sound source signals stored as representative speech units, thereby shortening the calculation time required for learning the synthesis unit. It is possible to reduce the amount of memory used in the rule synthesis system.

Moreover, since speech synthesis is performed using a combination of the LPC coefficient and the sound source signal (a _i , e _j ), the number of representative speech units of either the LPC coefficient or the number of sound source signals is determined by the number of clusters or the total number of synthesis units ( For example, even if it is smaller than the total number of CV and VC syllables), a good synthesized speech can be obtained.

In the sixth to ninth embodiments, a smoother synthesized speech can be obtained by considering the connection distortion between synthetic speech units as a distortion measure between the training speech unit and the synthesized speech unit. Can also.

Further, in the learning of the synthesis unit and the rule synthesis, the same adaptive post-filter as described in the second embodiment can be used together with the synthesis filter.
As a result, the spectrum of the synthesized speech is shaped, and a clear synthesized speech with a “slip” can be obtained.

[0101]

As described above, according to the speech synthesizing method of the present invention, at least one of the pitch and the duration is changed for the input speech unit, so that the level of the synthesized speech is natural. In order to evaluate the distortion of the voice and to select a speech unit selected from the input speech units based on the distortion as a representative speech unit, it is possible to generate a representative speech unit in consideration of the characteristics of the speech synthesizer. By connecting the representative speech units and performing speech synthesis, it is possible to generate high-quality synthesized speech close to natural speech.

Further, in the present invention, the spectrum synthesized for the speech synthesized by the connection of the representative speech units is further performed, and the similar spectrum shaping is also performed for the synthesized speech units, so that the speech after the spectrum shaping is obtained. At the level of the final synthesized speech signal, it is possible to generate a representative speech unit in which distortion with respect to natural speech is reduced, so that a clearer synthesized speech having a “slippery” can be generated.

Further, by selecting and connecting each representative speech unit in accordance with a unit selection rule based on the phoneme environment, the synthesized speech becomes smooth and highly natural. Furthermore, when storing information of a set of coefficients (for example, LPC coefficients) of a synthesis filter for generating a synthesized speech signal by inputting a sound source signal (for example, a prediction residual signal) as a representative speech unit, the input signal is quantized. As a result, the number of sound source signals stored as representative speech units and the number of synthesis filter coefficients can be reduced, so that the calculation time required for learning a synthesis unit can be reduced, and the amount of memory used in the rule synthesis system can be reduced. Can be reduced.

Further, the number of at least one of the sound source signal and the coefficient of the synthesis filter stored as the information of the representative speech unit is determined by the total number of speech synthesis units (for example, CV, VC
Even if it is smaller than the total number of syllables) or the number of phoneme environment clusters, a good synthesized speech can be obtained.

[Brief description of the drawings]

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

FIG. 2 is a flowchart showing a first processing procedure in a representative speech unit generation unit in FIG. 1;

FIG. 3 is a flowchart showing a second processing procedure in a representative speech unit generation unit in FIG. 1;

FIG. 4 is a flowchart showing a third processing procedure in a representative speech unit generation unit in FIG. 1;

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

FIG. 6 is a block diagram showing a configuration example of an adaptive post filter in FIG. 5;

FIG. 7 is a flowchart showing a first processing procedure in a representative speech unit generation unit in FIG. 5;

FIG. 8 is a flowchart showing a second processing procedure in the representative speech unit generation unit in FIG. 5;

FIG. 9 is a flowchart showing a third processing procedure in the representative speech unit generation unit in FIG. 5;

FIG. 10 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesis device according to a third embodiment of the present invention.

11 is a flowchart showing a processing procedure of the synthesis unit learning system of FIG.

FIG. 12 is a block diagram showing a configuration of a rule synthesis system in a speech synthesizer according to a third embodiment of the present invention.

FIG. 13 is a block diagram illustrating a configuration of a synthesis unit learning system in a speech synthesis device according to a fourth embodiment of the present invention.

FIG. 14 is a block diagram showing a configuration of a rule synthesis system in a speech synthesis device according to a fourth embodiment of the present invention.

FIG. 15 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesis device according to a fifth embodiment of the present invention.

16 is a flowchart showing a first processing procedure of the synthesis unit learning system of FIG.

FIG. 17 is a flowchart showing a second processing procedure of the synthesis unit learning system of FIG. 15;

FIG. 18 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesizer according to a sixth embodiment of the present invention.

FIG. 19 is a flowchart showing a processing procedure of the synthesis unit learning system of FIG. 18;

FIG. 20 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesizer according to a seventh embodiment of the present invention.

FIG. 21 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesis device according to an eighth embodiment of the present invention.

FIG. 22 is a block diagram showing a configuration of a synthesis unit learning system in a speech synthesis device according to a ninth embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Synthesis unit learning system 2 ... Rule synthesis system 11 ... Representative speech unit generation unit 12 ... Phoneme environment cluster storage unit 13 ... Representative speech unit storage unit 14 ... Unit selection unit 15 ... Speech synthesis unit 16 ... Adaptive post filter 21: Formant emphasis filter 22: Pitch emphasis filter 101: Training speech unit (first speech unit) 102: Phoneme environment labeled on the training speech unit 103: Input speech unit (second speech unit) 104 ... representative speech unit 105 ... phoneme environment cluster 106 ... phoneme environment cluster 107 ... representative speech unit selection information 108 ... representative speech unit 111 ... prosodic information 112 ... phoneme symbol string 113 ... synthesized speech signal 114 ... synthesized speech signal Reference Signs List 30 synthesis unit learning system 31 LPC analysis / inverse filter 31a LPC analysis unit 31b inverse filter 32 sound source signal storage 33 LPC coefficient storage unit 34 sound source signal generation unit 35 synthesis filter 36 distortion calculation unit 37 minimum distortion search unit 38 clustering unit 40 rule synthesis system 41 combination storage unit 42 sound source signal storage unit 43 LPC Coefficient storage unit 44 sound source signal generation unit 45 synthesis filter 48 clustering unit 51 phoneme environment cluster generation unit 52 cluster storage units 61 and 62 buffers 63 and 64 quantization table generation unit

Continuation of the front page (56) References JP-A-64-78300 (JP, A) JP-A-3-119394 (JP, A) JP-A-5-73100 (JP, A) JP-A-6-318094 (JP) , A) JP-A-8-263095 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 13/00-13/08 G10L 21/04

Claims (14)

(57) [Claims] An at least one of a pitch and a duration of each of a plurality of input speech units is set to at least one of a pitch and a duration of a plurality of training speech units that are the same as or different from the input speech unit. A plurality of synthesized speech units are generated by changing to be equal to one, and based on a distance measure between each of the synthesized speech units and each of the training speech units, Each of the distortions is evaluated, and a set of a predetermined number of speech units is selected from the input speech units based on a distortion evaluation function using the distortion information as a representative speech unit, and stored. A synthesized speech is generated by selecting and connecting a representative speech unit from the representative speech units according to an input phoneme. 2. A method according to claim 1, wherein at least one of the pitch and duration of each of the plurality of input speech units is any one of a plurality of training speech units which are the same as or different from the input speech units, and are labeled with a phoneme environment. A plurality of synthesized speech units are generated by changing them to be equal to at least one of the pitch and the duration length of the pitch, and the distance scale between each of the synthesized speech units and each of the training speech units Based on the input speech unit, a set of a predetermined number of speech units is used as a representative speech unit based on a distortion evaluation function using the information on the distortion. Selecting and storing, generating a plurality of phoneme environment clusters based on the clustering evaluation function using the information of the phoneme environment and the distortion, and generating that of the representative speech unit A speech is generated by selecting and connecting a representative speech unit corresponding to a phoneme environment cluster to which a phoneme environment of an input phoneme belongs from the representative speech units, and connecting the selected speech units. Synthesis method. 3. A training speech unit, wherein at least one of the pitch and duration of each of the plurality of input speech units is one of a plurality of training speech units which are the same as or different from the input speech units, and are labeled with a phoneme environment. A plurality of synthesized speech units are generated by changing them to be equal to at least one of the pitch and the duration length of the pitch, and the distance scale between each of the synthesized speech units and each of the training speech units Evaluating the distortion of each of the synthesized speech units based on the phoneme environment and a clustering evaluation function using the information of the distortion to generate and store a plurality of phoneme environment clusters, from the input speech unit A set of a predetermined number of speech units is selected as a representative speech unit based on the distortion evaluation function using the phoneme environment and the distortion information, and the phoneme is selected. A synthesized speech is generated by selecting and connecting a representative speech unit corresponding to a phoneme environment cluster to which a phoneme environment of an input phoneme belongs from the representative speech unit, and storing the synthesized speech. Speech synthesis method. 4. The method according to claim 1, wherein at least one of the pitch and the duration of each of the plurality of input speech units is at least one of the pitch and the duration of any of the plurality of training speech units identical to or different from the input speech unit. A plurality of synthesized speech units are generated by changing them to be equal to one, and the spectrum is shaped for the synthesized speech unit, and each of the synthesized speech units after the spectrum shaping is performed and the training speech unit. Evaluating the distortion of each of the synthesized speech units based on a distance measure between each of the segments, and a predetermined number of voices based on a distortion evaluation function using the information of the distortion from the input speech unit. Selecting and storing a set of segments as a representative speech unit, selecting and connecting a representative speech unit from the stored representative speech units according to an input phoneme; A synthesized speech is generated by performing a spectrum shaping on the synthesized speech to generate a final synthesized speech. 5. A method according to claim 1, wherein at least one of the pitch and the duration of each of the plurality of input speech units is any one of a plurality of training speech units that are the same as or different from the input speech units and are labeled with a phoneme environment. A plurality of synthesized speech units are generated by changing the synthesized speech units to be equal to at least one of the pitch and the duration of the synthesized speech unit, and the synthesized speech unit is subjected to spectrum shaping. Evaluating a distortion of each of the synthesized speech units based on a distance measure between each of the segments and each of the training speech units; and a distortion evaluation function using the information of the distortion from the input speech unit. A set of a predetermined number of speech units is selected and stored as a representative speech unit based on the above, and clustering using the information of the phoneme environment and the distortion is performed. A plurality of phoneme environment clusters are generated based on the evaluation function and stored in correspondence with each of the representative speech units. A representative phoneme corresponding to a phoneme environment cluster to which a phoneme environment of an input phoneme belongs from the representative speech unit A speech synthesis method comprising: generating a synthesized speech by selecting and connecting pieces; and performing spectrum shaping of the synthesized speech to generate a final synthesized speech. 6. A plurality of training speech units which are the same as or different from the input speech units and have at least one of a pitch and a duration length of each of the plurality of input speech units. A plurality of synthesized speech units are generated by changing the synthesized speech units to be equal to at least one of the pitch and the duration of the synthesized speech unit, and the synthesized speech unit is subjected to spectrum shaping. Evaluating the distortion of each of the synthesized speech units based on a distance measure between each of the segments and each of the training speech units, based on a clustering evaluation function using information of the phoneme environment and the distortion. Generate and store a plurality of phoneme environment clusters, and evaluate distortion from the input speech unit using information on the phoneme environment and the distortion Based on the function, a set of a predetermined number of speech units is selected as a representative speech unit and stored in correspondence with the phoneme environment cluster.From the representative speech unit to a phoneme environment cluster to which the phoneme environment of the input phoneme belongs. A speech synthesis method comprising: generating a synthesized speech by selecting and connecting corresponding representative speech units; and performing a spectrum shaping of the synthesized speech to generate a final synthesized speech. 7. The information processing apparatus according to claim 1, wherein information representing a set of a sound source signal and a set of coefficients of a synthesis filter for generating a synthesized speech signal by using the sound source signal as an input is generated and stored as the representative speech unit. 7. The speech synthesis method according to any one of 6. 8. The apparatus according to claim 1, wherein said sound source signal and coefficients of said synthesis filter are quantized, and information of a set of said quantized sound source signal and coefficients of said synthesis filter is generated and stored as said representative speech unit. The speech synthesis method according to claim 7. 9. Selective information of a set of a sound source signal and a set of coefficients of a synthesis filter for generating a synthesized speech signal using the sound source signal as an input is selected and stored as the representative speech unit; The number of at least one of the sound source signal and the coefficient of the synthesis filter stored as is less than the total number of speech synthesis units.
7. The speech synthesis method according to any one of 6. 10. As the representative speech unit, information of a set of a sound source signal and a coefficient set of a synthesis filter for generating a synthesized speech signal by inputting the sound source signal is selected and stored, and information of the representative speech unit is selected. The speech synthesis according to any one of claims 2, 3, 5, and 6, wherein the number of at least one of the sound source signal and the coefficient of the synthesis filter stored as is less than the total number of the phoneme environment clusters. Method. 11. A representative speech unit generation unit for generating a plurality of representative speech units, a representative speech unit storage unit for storing said representative speech units, and corresponding to an input phoneme from the stored representative speech units. A unit for selecting a representative speech unit to be connected, and a speech synthesis unit for generating a synthesized speech by connecting the selected representative speech units. At least one of the pitch and duration of each speech unit is changed to be equal to at least one of the pitch and duration of any of a plurality of training speech units that are the same as or different from the input speech unit. Thereby generating a plurality of synthesized speech units, and distorting each of the synthesized speech units based on a distance measure between each of the synthesized speech units and each of the training speech units. Evaluates, from the input speech segment, speech synthesis apparatus and selects a set of predetermined number of speech units as the representative speech units based on the evaluation function using the information of the strain. 12. A representative speech unit generation unit that generates a plurality of representative speech units and generates a plurality of phoneme environment clusters; a representative speech unit storage unit that stores the representative speech units; A phoneme environment cluster storage unit that stores a cluster; and a unit selection unit that selects a representative speech unit corresponding to a phoneme environment cluster to which a phoneme environment of an input phoneme belongs from the phoneme environment clusters from the representative speech unit. A voice synthesis unit that generates a synthesized voice by connecting the selected representative voice units, wherein the representative voice unit generation unit includes at least one of a pitch and a duration time of each of the plurality of input voice units. Is equal to at least one of the pitch and the duration of any of a plurality of training speech units that are the same as or different from the input speech unit and whose phoneme environment is labeled. A plurality of synthesized speech units are generated by changing so that each of the synthesized speech units is based on a distance scale between each of the synthesized speech units and each of the training speech units. Evaluating the distortion, from the input speech unit, selecting a set of a predetermined number of speech units as the representative speech unit based on a distortion evaluation function using the information of the distortion, A speech synthesizer for generating the phoneme environment cluster based on a classing evaluation function using information. 13. A representative speech unit generating unit for generating a plurality of representative speech units and a plurality of phoneme environment clusters; a representative speech unit storage unit for storing the representative speech units; A phoneme environment cluster storage unit for storing a cluster, and a unit for selecting a representative speech unit corresponding to a phoneme environment cluster to which the phoneme environment of the input phoneme belongs from the stored phoneme environment clusters, from the stored representative speech units. A selecting unit, and a voice synthesizing unit that generates a synthesized voice by connecting the selected representative voice units, wherein the representative voice unit generating unit includes a pitch and a duration time of each of the plurality of input voice units. At least one of the lengths is reduced in pitch and duration of any of a plurality of training speech units that are the same as or different from the input speech unit, and are labeled with a phoneme environment. Also, a plurality of synthesized speech units are generated by changing them to be equal to one, and the synthesized speech units are generated based on a distance measure between each of the synthesized speech units and each of the training speech units. Evaluating the respective distortions, generating the plurality of phoneme environment clusters based on the clustering evaluation function using the information on the phoneme environment and the distortion, and generating information on the phoneme environment and the distortion from the input speech unit. A speech synthesizer, wherein a set of a predetermined number of speech units is selected as the representative speech unit based on a distortion evaluation function using the above. 14. A post-filter for shaping the spectrum of the synthesized speech to generate a final synthesized speech, wherein the representative speech unit generation unit performs spectrum shaping on the synthesized speech unit, and The distortion of each of the synthesized speech units is evaluated based on a distance measure between each of the synthesized speech units after the shaping and each of the training speech units.
The speech synthesizer according to any one of claims 1 to 13.