JP2009133890A - Speech synthesis apparatus and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice synthesizing device capable of generating a natural synthesized voice of high tone quality. <P>SOLUTION: A voice synthesizing part 4 is composed of a voice element storage part 40, a phoneme environment storage part 41, a phoneme sequence-rhythm information input part 42, a voice element selecting-fusing part 43, and a voice element editing-connecting part 44. The voice element selecting-fusing part 43 is composed of a voice element selecting part 430, a voice element fusing part 431 and a phoneme environment computing part 432. An optimum element sequence is obtained in the voice element selecting part 430 using a phoneme environment parameter of the fused voice elements, and the phoneme environment parameter of the fused voice elements is obtained to synthesize voice. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to text-to-speech synthesis, and more particularly to a speech synthesis apparatus and method for generating a speech signal from information such as phoneme symbol strings, pitches, and phoneme durations.

  Artificially creating speech signals from arbitrary sentences is called “text-to-speech synthesis”. Text-to-speech synthesis is generally composed of three stages: a language processing unit, a prosody processing unit, and a speech synthesis unit.

  The input text is subjected to morphological analysis, syntax analysis, etc. in the language processing section as the first stage. Next, as a second step, the prosody processing unit performs accent and intonation processing, and outputs phoneme series / prosodic information (basic frequency, phoneme duration, power, etc.). After that, as a final step, the text signal synthesis is realized by synthesizing the voice signal from the phoneme sequence / prosodic information in the voice signal synthesis unit.

  The principle of a synthesizer capable of synthesizing such an arbitrary phoneme symbol string is as follows. When a vowel is represented by V and a consonant is represented by C, a characteristic parameter (speech element) of a basic small speech unit such as CV, CVC, VCV or the like. Is stored, and the voice is synthesized by controlling and connecting the pitch and duration. In this method, the stored speech segment greatly affects the quality of the synthesized speech.

  As one of such speech synthesis methods, a plurality of speech units are selected for each speech unit from a large number of speech units, with the target phoneme sequence / prosodic information as a target. There is a multi-speech unit selection / fusion type speech synthesis method in which new speech units are generated by merging and synthesizing speech by connecting them (for example, see Patent Document 1).

  In this multi-speech unit selection / fusion type speech synthesis, a speech unit is first selected from a large number of pre-stored speech units based on the input phoneme sequence / prosodic information. As a speech unit selection method, there is a method in which the degree of distortion of synthesized speech generated by synthesizing speech is defined as a cost function and a speech unit is selected so as to reduce the cost.

For example, the target distortion representing the difference between the target speech and each speech segment, such as the prosody and phonological environment, and the connection distortion generated by connecting speech segments are quantified as costs. Based on this cost, a speech segment to be used for speech synthesis is selected. Further, for example, the pitch waveforms are averaged, or the centroids of a plurality of selected speech segments are used for fusion. As a result, it is possible to suppress deterioration of sound quality in editing and connection of speech units and to obtain a stable synthesized speech.
JP 2005-164749

  In the multi-speech unit selection / fusion type speech synthesis method as described above, when selecting a fused speech unit candidate, a speech unit is not considered in advance without considering connection distortion with a speech unit used as synthesized speech. In other words, because the fusion speech unit candidate is obtained based on the connection distortion with the speech unit on the optimal unit sequence that is not actually used as the speech unit of the synthesized speech, There is a problem that discontinuity occurs in the connection portion of the generated synthesized speech.

  Therefore, in view of the above problems, the present invention provides a speech synthesizer and method for generating a more natural and high-quality synthesized speech in a multiple speech unit selection / fusion speech synthesis. Objective.

  The present invention divides a speech unit group, a storage unit storing phoneme environment parameters for each speech unit of the speech unit group, and a phoneme sequence corresponding to a target speech to be synthesized into synthesis units. For each of the plurality of segments obtained by the above, a selection unit that selects a plurality of first speech units from the speech unit group and a plurality of the first speech units are combined to form one second speech. Synthesis by connecting a generation unit that generates a segment, a parameter calculation unit that calculates a phoneme environment parameter of the second speech unit, and the second speech unit generated for each of the segments A segment setting unit configured to set, as the segment of interest, one segment for selecting the first speech segment from the segments. An extraction unit that extracts a plurality of speech units having the same characteristics as the phoneme of the target segment from the speech unit group as speech unit candidates, and uses each speech unit candidate of the target segment. A first cost calculation unit that calculates a target cost representing a distortion amount of the synthesized speech generated from the phoneme environment parameter of each speech unit candidate and the prosodic information of the target speech; Alternatively, each connection cost that represents the amount of distortion that occurs when the speech unit of an adjacent segment that is an adjacent segment and the speech unit candidate of the target segment are connected is calculated. ) When the adjacent segment is the second speech unit, from the phoneme environment parameter of the second speech unit and the phoneme environment parameter of each speech unit candidate Or (2) if the adjacent segment does not have the second speech segment, the phoneme environment parameter of the speech segment candidate corresponding to the segment of interest and the adjacent segment A second cost calculation unit that calculates the connection cost from the phoneme environment parameter of the speech unit candidate corresponding to the target speech unit, and the target cost and the connection cost among the plurality of speech unit candidates in the segment of interest. A speech unit selection unit that selects a plurality of speech unit candidates having a low total cost as the first speech unit.

  The present invention also provides a speech unit group, a storage unit storing phoneme environment parameters for each speech unit of the speech unit group, and a phoneme sequence corresponding to a target speech to be synthesized in synthesis units. A segment setting unit for setting one segment as a target segment from among a plurality of segments obtained by dividing, and a plurality of speech units having the same characteristics as the phonemes of the target segment from the speech unit group Are extracted as a third speech unit, a generation unit that generates a fourth speech unit by fusing the plurality of third speech units, and a phoneme environment parameter of the fourth speech unit. A target cost representing a distortion amount of the synthesized speech generated using the parameter calculation unit to calculate, and the third speech unit and the fourth speech unit of the segment of interest, A third cost calculation unit for calculating each of the third speech unit and the fourth speech unit from the phoneme environment parameters and the prosody information of the target speech, and an adjacent segment that is adjacent to the target segment Each connection cost representing the amount of distortion generated when the third speech unit and the fourth speech unit of the segment and the third speech unit and the fourth speech unit of the segment of interest are connected. Is calculated from the phoneme environment parameters of the third speech unit and the fourth speech unit of the segment of interest and the phoneme environment parameters of the third speech unit and the fourth speech unit of the adjacent segment. The total cost of the target cost and the connection cost is low among the four cost calculation unit and the plurality of third speech units and the fourth speech unit in the segment of interest. A synthesized speech is generated by connecting a speech unit selection unit that selects a plurality of speech units as the fifth speech unit and the fifth speech unit generated for each of the segments. And a synthesis unit.

  According to the present invention, it is possible to provide a speech synthesizing method that generates a more natural and high-quality synthesized speech by reducing the degree of deterioration of the sound quality as compared with a case where connection distortion with a fusion speech unit is not taken into consideration.

  A speech synthesizer that performs text-to-speech synthesis according to an embodiment of the present invention will be described with reference to the drawings.

(First embodiment)
A speech synthesizer according to a first embodiment of the present invention will be described with reference to FIGS.

(1) Configuration of Speech Synthesizer FIG. 1 is a block diagram showing a configuration of a speech synthesizer according to this embodiment.

  As shown in FIG. 1, the speech synthesizer includes a text input unit 1, a language processing unit 2, a prosody processing unit 3, a speech synthesis unit 4, and a speech waveform output unit 5.

  The function of each part can also be realized by a program stored in a computer.

  The language processing unit 2 performs morphological analysis / syntax analysis of the text input from the text input unit 1 and sends the result to the prosody processing unit 3.

  The prosody processing unit 3 performs accent and intonation processing from the language analysis result, generates a phoneme sequence (phoneme symbol string) and prosody information, and sends them to the speech synthesis unit 4.

  The speech synthesizer 4 generates a speech waveform from the phoneme sequence and prosodic information.

  The voice waveform output unit 5 outputs the voice waveform thus generated.

(2) Configuration of Speech Synthesizer 4 FIG. 2 is a block diagram showing a configuration example of the speech synthesizer 4 in FIG.

  As shown in FIG. 2, the speech synthesis unit 4 includes a speech unit storage unit 40, a phoneme environment storage unit 41, a phoneme sequence / prosodic information input unit 42, a speech unit selection / fusion unit 43, and a speech unit editing / connection unit. 44.

  Hereinafter, the function of each part 40-44 is demonstrated in detail.

(3) Speech unit storage unit 40
A large amount of speech units are accumulated in the speech unit storage unit 40, and speech units of speech units (synthesis units) used when generating synthesized speech are stored.

  A synthesis unit is a phoneme or a combination of phonemes. For example, semiphonemes, phonemes (C, V), diphones (CV, VC, VV), triphones (CVC, VCV), syllables (CV, V), etc. There may be. V represents a vowel, and C represents a consonant.

  In addition, the speech segment represents a waveform of a speech signal corresponding to a synthesis unit or a parameter series representing its characteristics.

  In the speech unit storage unit 40, for example, when the speech unit is a phoneme, as shown in FIG. 4, the waveform of the speech signal of each phoneme is stored together with the speech unit number for identifying the phoneme. Yes. Each speech unit stored in the speech unit storage unit 40 is labeled for each phoneme with respect to a large number of separately collected speech data, and a speech waveform cut out for each phoneme is used as a speech unit. Accumulated.

(4) Phoneme environment storage unit 41
The phoneme environment storage unit 41 stores phoneme environment parameters of speech units stored in the speech unit storage unit 40.

  The phoneme environment parameter of the speech segment is information corresponding to a combination of factors that become the phoneme environment for the speech segment. Factors include, for example, the phoneme name of the speech unit, the preceding phoneme, the succeeding phoneme, the succeeding phoneme, the fundamental frequency, the phoneme duration, power, the presence or absence of stress, the position from the accent core, the time from breathing, the utterance There are speed, feelings, etc.

  In the phoneme environment storage unit 41, for example, when the speech unit is a phoneme, as shown in FIG. 5, the phoneme environment parameters of each phoneme stored in the phoneme unit storage unit 40 include the phoneme speech unit. It is stored in association with a single number. Here, phoneme symbols (phoneme names), fundamental frequencies, phoneme duration durations, and cepstrum coefficients at both ends of a speech unit are stored as phoneme environment parameters.

(5) Phoneme sequence / prosodic information input unit 42
The phoneme sequence / prosodic information input unit 42 receives the phoneme sequence and prosodic information of the target speech output from the prosody processing unit 3. The prosodic information input to the phoneme sequence / prosodic information input unit 42 includes a fundamental frequency, a phoneme duration, power, and the like.

  Hereinafter, the phoneme sequence and the prosody information input to the phoneme sequence / prosodic information input unit 42 are referred to as “input phoneme sequence” and “input prosody information”, respectively. The input phoneme sequence is a sequence of phoneme symbols, for example.

  In the phoneme sequence / prosodic information input unit 42, for example, if the speech segment is a phoneme, the phoneme information is subjected to morpheme analysis / syntax analysis of the input text for text-to-speech synthesis, and further performs accent and intonation processing. The prosodic information and phoneme sequence obtained in this way are input. The input prosody information includes a fundamental frequency and a phoneme duration.

(6) Speech unit selection / fusion unit 43
Next, the speech element selection / fusion unit 43 will be described.

  FIG. 3 is a block diagram illustrating a configuration example of the speech unit selection / fusion unit 43 of FIG.

  As shown in FIG. 3, the speech unit selection / fusion unit 43 includes a speech unit selection unit 430, a speech unit fusion unit 431, and a phoneme environment calculation unit 432.

(6-1) Speech unit selection unit 430
The speech unit selection unit 430, for each of a plurality of segments obtained by dividing the input phoneme sequence by synthesis unit, prosody information included in the phoneme environment parameter of the speech unit or the phoneme described later. It is stored in the speech unit storage unit 40 so as to estimate the amount of distortion that is the degree of distortion with the phoneme environment parameter of the fused speech unit obtained by the environment calculation unit 432 and to minimize the amount of distortion. A speech unit is selected from existing speech units.

  As the amount of distortion, a cost function described later can be used, but is not limited to this.

(6-2) Speech unit fusion unit 431
The speech unit fusion unit 431 merges a plurality of speech units selected by the speech unit selection unit 430 to generate a new speech unit.

(6-3) Phoneme environment calculation unit 432
The phoneme environment calculation unit 432 calculates the phoneme environment parameters of the speech units fused by the speech unit fusion unit 431. By performing this operation for each segment, a new speech segment sequence corresponding to the phoneme symbol sequence of the input phoneme sequence is obtained.

(7) Speech segment editing / connection unit 44
In the speech segment editing / connection unit 44, a new speech segment sequence is transformed and connected based on the input prosodic information, and a speech waveform of a synthesized speech is generated.

  The speech waveform generated in this way is output by the speech waveform output unit 5 of FIG.

(8) Processing Contents of Speech Unit Selection / Fusion Unit 43 Next, the flow of processing in the speech unit selection / fusion unit 43 will be described with reference to FIG. Here, it is assumed that the speech unit of the synthesis unit is a phoneme. FIG. 6 is a flowchart showing the flow of processing in the speech unit selection / fusion unit 43.

  In the present embodiment, the number of segments of the synthesized speech is I, and speech units are merged from the beginning of the sentence toward the end of the sentence (that is, in time series).

  Further, Step S4300, Step S4310, Step S4320, and Step S4330 are repeated I times, and processing is performed so that I segments become the target segment once. Hereinafter, each step will be described.

(9) Step S4300
First, in step S4300, an optimum segment sequence is obtained based on a cost function described later.

(9-1) Cost function The cost function is determined as follows.

First, sub-cost functions C _n (u _i, u _i−1 , t _i ) (n: 1,..., N for each factor of distortion generated when speech units are deformed and connected to generate synthesized speech. , N defines the number of sub-cost functions). Here, t _i corresponds to the i-th segment when the target speech (target speech) corresponding to the input phoneme sequence and the input prosodic information is t = (t ₁ ,..., T _I ). The target phoneme environment parameter information of the partial speech unit is represented, and u _i represents the speech unit having the same phoneme as t _i among the speech units stored in the speech unit storage unit 40.

  The sub-cost function is used to calculate a cost for estimating the amount of distortion of the synthesized speech with respect to the target speech that occurs when the synthesized speech is generated using the speech units stored in the speech unit storage unit 40. It is.

  In order to calculate the cost, here, specifically, a target cost for estimating a distortion amount of the synthesized speech generated by using the speech unit with respect to the target speech, and the speech unit is used as another speech unit. Two types of sub-costs are used which are connection costs for estimating the amount of distortion of the synthesized speech that occurs when connected to a piece with respect to the target speech.

  As the “target cost”, the fundamental frequency cost representing the difference (difference) between the fundamental frequency of the speech element stored in the speech element storage unit 40 and the target fundamental frequency, the phoneme duration length of the speech element, A phoneme duration cost that represents a difference (difference) from the target phoneme duration is used.

  As the “connection cost”, a spectrum connection cost representing a difference (difference) in spectrum at the connection boundary is used.

Specifically, the fundamental frequency cost is

Calculate from Here, v _i is the phoneme environment parameters of speech unit u _i stored in the voice unit storage 40, f represents a function to extract the fundamental frequency from the phonetic environment parameter v _i.

Also, the long phoneme duration cost is

Calculate from Here, g represents the function to extract phoneme duration from the phonetic environment parameter v _i.

Also, the spectrum connection cost is the cepstrum distance between two speech segments:

Calculate from Here, h represents a function for taking out a cepstrum coefficient of a connection boundary of the speech unit u _i as a vector.

Define the weighted sum of these subcost functions as the composite unit cost function:

Here, w _n represents the weight of the sub cost function. In the present embodiment, for the sake of simplicity, all w _n is set to "1". The above formula (4) is the synthesis unit cost of the speech unit when a speech unit is applied to a synthesis unit.

For each of a plurality of segments obtained by dividing the input phoneme sequence by the synthesis unit, the result of calculating the synthesis unit cost from the above equation (4) is the sum of all segments is called the cost. A cost function for calculation is defined as shown in the following equation (5):

(9-2) Optimal Segment Sequence In step S4300 in FIG. 6, one speech segment is selected per segment (ie, per synthesis unit) using the cost functions shown in the above equations (1) to (5). By using this, a sequence of speech units having a minimum cost value calculated by the above equation (5) is obtained.

  Further, for a segment for which there is no fused speech unit generated by the speech unit fusion unit 431 described later, one of the speech units in the speech unit storage unit 40 is used, and the above equation (5) is used. ) To obtain a speech unit sequence having the smallest cost value.

  Furthermore, for the segment in which the fusion speech unit generated by the speech unit fusion unit 431 exists, the fusion speech unit is used, and the speech unit having the smallest cost value calculated by the above equation (5) is used. Find the series.

  A combination of speech units that minimizes the cost is referred to as an “optimal unit sequence”. That is, each speech unit in the optimum unit sequence corresponds to each of a plurality of segments obtained by dividing the input phoneme sequence by a synthesis unit, and is calculated from each speech unit in the optimum unit sequence. The cost value calculated from the synthesis unit cost and the equation (5) is smaller than any other speech unit sequence. It should be noted that the search for the optimum unit sequence can be performed more efficiently by using dynamic programming (DP).

(9-3) Specific Example For example, as shown in FIG. 7, it is assumed that the input phoneme sequence is “ts · i · i · s · a ·. In this case, the synthesis unit corresponds to each of the phonemes “ts”, “i”, “i”, “s”, “a”,..., And each of these phonemes corresponds to one segment.

  A segment corresponding to the third phoneme “i” in the input phoneme sequence is set as a target segment. That is, the first phoneme “ts” and the second phoneme “i” in the input phoneme sequence are already speech unit fused in step S4320, and the phoneme environment parameters of the speech units fused in step S4330 are calculated. Has been.

  In this case, speech units 4301a and 4301b corresponding to the first phoneme “ts” and the second “i” on the optimal unit sequence are fused speech units, and the remaining phonemes in the phoneme sequence are the same. In the third phoneme “i”, the fourth phoneme “s”, and the fifth phoneme “a”,..., 4301c, 4301d, 4301e,. It is selected as a speech segment on the series 4301.

  When a speech unit sequence consisting of speech units 4301a, 4301b, 4301c, 4301d, 4301e,... On this optimal unit sequence 4301 is used, the cost calculated from the synthesis unit cost and the equation (5) is different. The value is smaller than the speech unit sequence of the throat.

(10) Step S4310
Next, proceeding to step S4310, a plurality of speech segments are selected per segment using the optimum segment sequence obtained at step S4300. Here, a description will be given assuming that M speech segments are selected for each of I segments. Details are shown in the flowchart of FIG.

  In step S4311, ranking is performed according to the cost value calculated by equation (5), and in step S4312, the top M speech segments are selected.

  For example, as in FIG. 7, in FIG. 9, it is assumed that the input phoneme sequence is “ts · i · i · s · a ·. FIG. 9 shows a case where a segment corresponding to the third phoneme “i” in the input phoneme sequence is set as a target segment, and a plurality of speech segments are obtained for this target segment. For segments other than the segment corresponding to the third phoneme “i”, the speech units 4313a, 4313b, 4313d, 4313e,. Compared with the speech unit in the optimum sequence of FIG. 7, the speech unit 4313a is the fused speech unit 4301a, the speech unit 4313b is the fused speech unit 4301b, the speech unit 4313d is the speech unit 4301d, The segment 4313e corresponds to the speech segment 4301e.

  In this state, among the speech elements stored in the speech element storage unit 40, for each speech element having the same phoneme name (phoneme symbol) as the phoneme “i” of the segment of interest, Equation (5) is obtained. To calculate the cost. However, when the cost is calculated for each speech unit, the value changes because of the target cost of the target segment, the connection cost between the target segment and the previous segment, the target segment and the next one Since these are the connection costs with the segments, only these costs need be considered.

  That is, the procedure is as follows.

(Procedure 1) Among the speech elements stored in the speech element storage unit 40, one of the speech elements having the same phoneme name (phoneme symbol) as the phoneme “i” of the segment of interest is selected as the speech element. and u _3. From the fundamental frequency f (v ₃ ) of the speech unit u ₃ and the target fundamental frequency f (t ₃ ), the fundamental frequency cost is calculated using Equation (1).

(Procedure 2) The phoneme duration cost is calculated from the phoneme duration g (v ₃ ) of the speech unit u ₃ and the target phoneme duration g (t ₃ ) using Equation (2). To do.

Since the (Step 3) cepstral coefficients of the speech unit _{u 3} h _{(u 3),} and fused speech unit 4313b cepstrum coefficients h _{(u 2)} of the _{(u 2),} using equation (3), the The spectrum connection cost of 1 is calculated. Further, since the cepstrum coefficient of the speech unit _{u 3} h _{(u 3),} and cepstral coefficients of the fused speech unit _{4313d (u 4) h (u} 4), using equation (3), the second Calculate the spectrum connection cost.

(Procedure 4) A speech unit is calculated by calculating a weighted sum of the fundamental frequency cost, the phoneme duration time cost, and the first and second spectrum connection costs calculated by using each sub-cost function in the above-described Procedure 1 to Procedure 3. to calculate the cost of u _3.

(Procedure 5) For each speech unit having the same phoneme name (phoneme symbol) as the phoneme “i” of the segment of interest among the speech units stored in the speech unit storage unit 40, the above steps 1 to 4 are performed. When the cost is calculated according to the above, ranking is performed so that the speech unit having the smallest value has a higher rank (step S4311 in FIG. 8). Then, the top M speech segments are selected (step S4312 in FIG. 8). For example, in FIG. 9, the speech unit 4314a has the highest rank, and the speech unit 4314d has the lowest rank.

  The above steps 1 to 5 are performed for each segment. As a result, M speech segments are obtained for each segment.

  The phoneme environment parameters have been described as information on the phoneme of the speech unit and its fundamental frequency and phoneme duration, but are not limited to these, and the phoneme, the fundamental frequency, the phoneme duration, the preceding phoneme, if necessary. , Subsequent phonemes, subsequent phonemes, power, presence / absence of stress, position from the accent core, time from breathing, generation speed, emotion, etc. can be used in combination.

(11) Step S4320
Next, the process of step S4320 in FIG. 6 will be described.

  In step S4320, for the segment of interest, the M speech units are fused for each segment from the M speech units obtained in step S4310, and a new speech unit (fused speech unit) is created. Is generated. Although the waveform of voiced sound has a period, the waveform of unvoiced sound does not have a period, so this step performs different processing depending on whether the speech segment is voiced sound or unvoiced sound.

  First, the case of voiced sound will be described. In the case of voiced sound, a pitch waveform is extracted from the speech segment and fused at the level of the pitch waveform to create a new pitch waveform. A pitch waveform is a relatively short waveform that has a length up to several times the fundamental period of the speech and does not have a fundamental period, and whose spectrum represents the spectral envelope of the speech signal. means.

  As extraction methods, a pitch waveform is obtained by simply cutting out with a fundamental period synchronization window, a method of performing inverse discrete Fourier transform on a power spectrum envelope obtained by cepstrum analysis or PSE analysis, and an impulse response of a filter obtained by linear prediction analysis. There are various methods such as a method for obtaining, and a method for obtaining a pitch waveform that reduces distortion with respect to natural speech at the level of synthesized speech by a closed loop learning method.

  In the present embodiment, an example in which a pitch waveform is extracted using a method of cutting out with a basic period synchronization window will be described with reference to the flowchart of FIG. Here, a processing procedure in the case where one new speech unit is generated by fusing M speech units with respect to one segment among a plurality of segments will be described.

(11-1) Step S4321
First, in step S4321, a mark (pitch mark) is added to each speech waveform of the M speech units for each periodic interval.

  FIG. 11 (a) shows a case where pitch marks 4321b are attached to the speech waveform 4321a of one speech unit among the M speech units at every periodic interval.

(11-2) Step S4322
Next, in step S4322, as shown in FIG. 11B, windowing is performed with the pitch mark as a reference to cut out the pitch waveform.

  A Hanning window 4321c is used as the window, and the window length is twice the basic period.

  Then, as shown in FIG. 11C, the windowed waveform 4321d is cut out as a pitch waveform. Each of the M speech segments is subjected to the process as shown in FIG. 11 (the process of step S4322).

  As a result, for each of the M speech units, a series of pitch waveforms composed of a plurality of pitch waveforms is obtained.

(11-3) Step S4323
Next, proceeding to step S4323, the pitches in the series of all M pitch waveforms are matched to the one having the largest number of pitch waveforms among the pitch waveform series of the M speech units of the segment. The pitch waveforms are duplicated so that the number of pitch waveforms is the same (for a series of pitch waveforms with a small number of pitch waveforms).

  FIGS. 12A and 12B show pitch waveform series e1 to e3 cut out in step S4322 from each of the M segment (for example, three in this case) speech segments d1 to d3. ing. Since the number of pitch waveforms in the pitch waveform series e1 is 7, the number of pitch waveforms in the pitch waveform series e2 is 5, and the number of pitch waveforms in the pitch waveform series e3 is 6, the pitch waveform. Among the series e1 to e3, the series e1 has the largest number of pitch waveforms.

  Therefore, in FIG. 12C, according to the number of pitch waveforms in the series e1 (for example, the number of pitch waveforms is 7 here), the other series e2 and e3 are respectively the series. A state is shown in which one of the inside pitch waveforms is copied to make the number of pitch waveforms seven.

  As a result, new pitch waveform series corresponding to e2 and e3 are e2 ′ and e3 ′, respectively.

(11-4) Step S4324
Next, it progresses to step S4324. In this step, processing is performed for each pitch waveform.

  In step S4324, the pitch waveforms corresponding to the M speech units of the segment are averaged for each position, and a new pitch waveform sequence is generated. The generated new pitch waveform sequence is used as a fused speech unit.

  In FIG. 12 (d), the first to seventh pitch waveforms are averaged by three speech segments, respectively, to generate a new pitch waveform series f1 composed of seven new pitch waveforms. For example, the centroid of the first pitch waveform of the series e1, the first pitch waveform of the series e2 ′, and the first pitch waveform of the series e3 ′ is obtained and set as a new first pitch waveform. . The same applies to the second to seventh pitch waveforms of the new pitch waveform series f1. The series f1 of pitch waveforms is the “fused speech segment”.

  In addition, although it matched with what has the most pitch waveform among M speech units, you may match with the number of pitch marks of the synthetic speech unit to produce.

  Further, when the pitch waveforms are merged, the centroids are merged, but the present invention is not limited to this. For example, a method may be used in which an average of M pitch waveforms is obtained, an M speech element is divided into bands, and an average is obtained after aligning phases in each band.

(11-5) In the case of an unvoiced sound segment On the other hand, in the process of step S4320 in FIG. 6, in the case of an unvoiced sound segment, among the M speech elements of the segment in the speech element selection step S4310, the M The speech unit having the first rank assigned to each speech unit is used as it is.

(12) Step S4330
Next, the process of step S4330 in FIG. 6 will be described.

  In step S4330, the phoneme environment parameter of the fused speech unit obtained in step S4320 is calculated. The flow of processing in step S4330 in FIG. 6 is shown in the flowchart in FIG.

  The phoneme environment parameter of the fused speech unit is used when obtaining the optimum unit sequence in step S4300 in FIG.

  Therefore, in step S4331, the fundamental frequency of the fused speech unit is obtained.

  In step S4332, the phoneme duration duration of the fused speech segment is obtained.

  In step S4333, the phoneme environment parameters of the fused speech units are obtained by obtaining the cepstrum coefficient vector of the connection boundary of the fused speech units.

  Here, the fundamental frequency, phoneme duration, and cepstrum of the connection boundary of the united speech unit are obtained, but the present invention is not limited to this. It can also be changed according to phoneme environment parameters required for cost calculation.

  As described above, for each of a plurality of segments corresponding to the input phoneme sequence, the M speech units are merged from the M speech units selected for the segment, and a new speech unit is obtained. If (the fused speech unit) is generated, the process proceeds to the fused speech unit editing / connection step S4340 in FIG.

(13) Step S4340
In step S4340, the speech unit editing / connecting unit 44 generates a speech waveform (synthesized speech) by transforming and connecting the speech units merged for each segment obtained in step S4320 according to the input prosodic information. To do.

  Since the fused speech unit obtained in step S4320 is actually in the form of a pitch waveform, the fundamental frequency and the phoneme duration length of the fused speech unit are indicated in the input prosodic information. The speech waveform can be generated by superimposing the pitch waveform so as to be the basic frequency of the target speech and the phoneme duration of the target speech.

  FIG. 14 is a diagram for explaining the processing in step S4340. In FIG. 14, the fused speech segments obtained in step S4320 for each synthesis unit of phonemes “o”, “N”, “s”, “e”, and “N” are transformed and connected, 'Is generated. As shown in FIG. 14, according to the target fundamental frequency and the target phoneme duration length indicated in the input prosodic information, each pitch waveform in the united speech unit is segmented for each segment (synthesis unit). Change the basic frequency (change the pitch) or increase the number of pitch waveforms (change the time length). After that, synthesized speech is generated by connecting adjacent pitch waveforms within and between segments.

(14) Sub-cost requirement The target cost is to change the basic frequency of the fused speech unit as described above or the phoneme duration length based on the input prosodic information in order to generate synthesized speech. It is desirable to estimate (evaluate) as accurately as possible the distortion of the synthesized speech caused by the above with respect to the target speech. The target cost calculated from the equations (1) and (2), which are examples of such a target cost, is obtained by calculating the distortion amount from the speech element stored in the prosodic information of the target speech and the speech unit storage unit 40. It is calculated based on the difference between pieces of prosodic information.

  The connection cost is to estimate (evaluate) the distortion of the synthesized speech with respect to the target speech as accurately as possible as a result of connecting the fused speech segments as described above to generate the synthesized speech. Is desirable. The connection cost calculated from Expression (3), which is an example of such a connection cost, is fused with the speech unit stored in the speech unit storage unit 40 or in the speech unit fusion step S4320. It is calculated based on the difference in cepstrum coefficients at the connection boundary with the speech element.

(15) Effects The effects of this embodiment will be described while explaining the difference between the speech synthesis method according to the present embodiment and the conventional multiple speech unit selection / fusion speech synthesis method.

  In the speech synthesizer shown in FIG. 3 according to this embodiment, there is a phoneme environment calculation unit 432, which is optimal in the processing step S4300 in the speech unit selection unit 430 using the phoneme environment parameters of the fused speech units. It differs from the conventional speech synthesizer (see, for example, Patent Document 1) in that the segment series is obtained again and the phoneme environment parameter of the speech unit fused in step S4330 is obtained.

  In this embodiment, when selecting a speech unit, if adjacent segments have already been selected and a fused speech unit exists, the connection distortion with the fused speech unit is considered. By selecting and merging speech segments, it is possible to create speech segments with higher sound quality by eliminating discontinuities between connections, resulting in more natural and higher-quality synthesized speech. be able to.

(Second Embodiment)
Next, the speech element selection unit 430 according to the second embodiment will be described with reference to FIG.

  In the optimum unit sequence 4301 of FIG. 7, in the segment where the fused speech unit exists, the united speech unit is used as a speech unit on the optimum unit sequence. It is not limited to. Therefore, the second embodiment will be described as an example of this change.

  The synthesis unit cost in the optimum segment sequence 4301 and the cost calculated from the equation (5) are calculated as follows.

  As shown in FIG. 15, the speech unit on the optimal unit sequence is calculated from the speech unit storage unit 40 by the synthesis unit cost and the equation (5) regardless of the presence or absence of the fused speech unit. Consider a case where the cost is lower than the cost in the optimum segment sequence 4302 selected to be the minimum. In this case, the connection cost in each segment is calculated using the optimum segment sequence 4302.

  As a result, compared to the first embodiment, it is possible to stably generate a synthesized speech with a more natural and higher sound quality.

(Third embodiment)
Next, the speech element selection / fusion unit 43 according to the third embodiment will be described with reference to FIG.

  In FIG. 6, the speech units are merged from the beginning of the sentence toward the end of the sentence (that is, according to the time series), but the present invention is not limited to this. Therefore, a third embodiment will be described as this modification.

  When it is desired to particularly improve the quality of the synthesized speech of a specific word or sentence end part in the phoneme sequence at the time of synthetic speech generation, the segment corresponding to the specific word or sentence end part can be fused first.

  FIG. 16 is a flowchart showing the processing of the speech unit selection / fusion unit 43 according to this embodiment.

In step S4350, the order O _i (i: 1,..., I, I is the number of segments) set by the user who uses the synthesized speech method of the present embodiment is set. The O _i are applied from i = 1 one by one number until 1~I for each segment in order, steps for _{O i} th segment S4300, step S4310, step S4320, the processing in step S4330 Then, the same processing is repeated for the O _{i +} 1th segment.

  Thereby, compared to the first embodiment, the degree of freedom of speech segment selection in a segment in which the user particularly wants to improve the quality of synthesized speech, such as a specific word or the end of a sentence, is improved. As a result, the user Can generate synthesized speech with higher natural quality than desired.

(Fourth embodiment)
Next, the speech element selection / fusion unit 43 according to the fourth embodiment will be described with reference to FIG.

  In step S4310 of FIG. 6, M speech units are selected for each segment, but the present invention is not limited to this. Therefore, the fourth embodiment will be described as this modification.

A speech unit generated by merging L (<M) speech units from speech units stored in the speech unit storage unit 40 without merging speech units used when generating synthesized speech. Can also be used. In other words, the speech unit with the same phoneme as the phoneme of a segment, when the voice unit storage 40 is L _i number, per segment,

  This means that there are speech units as speech unit selection candidates.

  FIG. 17 is a flowchart showing the flow of processing in the speech unit selection / fusion unit 43 in FIG.

  In step S4320, m speech elements are fused, and in step S4360, phoneme environment parameters of the speech elements obtained by fusing m speech elements are calculated. This is repeated M times to generate speech segment candidates in one segment. Further, step S4320 and step S4360 are similarly repeated in each segment.

  Next, in step S4300, an optimal segment sequence is searched for speech segment candidates of each segment, and the speech segment selected for each segment is used as synthesized speech.

  As a result, compared to the first embodiment, it is possible to generate a speech unit of better quality from speech units stored in the speech unit storage unit 40, and a more natural and higher-quality synthesis. Can generate audio.

(Example of change)
Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage.

  In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

It is a block diagram which shows the structure of the speech synthesizer which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structural example of a speech synthesizer. It is a block diagram which shows the structural example of a speech unit selection / fusion | melting part. It is a figure which shows the example of a memory | storage of the speech unit of a speech unit storage part. It is a figure which shows the memory example of the phoneme environment parameter of a phoneme environment memory | storage part. It is a flowchart for demonstrating the processing operation | movement of a speech unit selection / fusion | melting part. It is a schematic diagram which shows the example of an optimal segment series. It is a flowchart for demonstrating the process of a speech unit selection part. It is a flowchart for demonstrating the process of a speech unit selection part. It is a flowchart for demonstrating the process of a speech unit fusion part. It is a flowchart for demonstrating the process which cuts out a pitch waveform. It is a flowchart for demonstrating the process which arranges the number of pitch waveforms. It is a flowchart for demonstrating the process of a phoneme environment calculation part. It is a figure for demonstrating the processing operation | movement of a speech unit edit and a connection part. It is a schematic diagram which shows the example of the optimal segment series which concerns on 2nd Embodiment. It is a flowchart for demonstrating the processing operation of the speech unit selection / fusion part which concerns on 3rd Embodiment. It is a flowchart for demonstrating the processing operation of the speech unit selection / fusion unit according to the fourth embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Text input part 2 Language processing part 3 Prosody processing part 4 Speech synthesis part 5 Speech waveform output part 40 Speech segment memory | storage part 41 Phoneme environment parameter memory | storage part 42 Phoneme series / prosodic information input part 43 Speech segment selection / fusion part 44 Speech segment editing / connection

Claims (18)

A storage unit storing phoneme environment parameters for each speech unit of the speech unit group and the speech unit group;
A selection unit for selecting a plurality of first speech units from the speech unit group for each of a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
A generating unit that generates one second speech unit by fusing the plurality of first speech units;
A parameter calculation unit for calculating a phoneme environment parameter of the second speech unit;
A synthesis unit that generates synthesized speech by connecting the second speech units generated for each of the segments;
Have
The selection unit includes:
A segment setting unit configured to set one segment for selecting the first speech segment from among the segments as a target segment;
An extraction unit that extracts a plurality of speech units having the same characteristics as the phoneme of the segment of interest from the speech unit group, as speech unit candidates;
A target cost representing a distortion amount of the synthesized speech generated using each speech unit candidate of the segment of interest is calculated from the phoneme environment parameter of each speech unit candidate and the prosodic information of the target speech. A first cost calculator that
Calculating each connection cost representing the amount of distortion that occurs when a speech unit of an adjacent segment, which is a segment adjacent before or after the target segment, and each speech unit candidate of the target segment are connected; (1) When the adjacent segment is the second speech unit, the connection cost is calculated from the phoneme environment parameter of the second speech unit and the phoneme environment parameter of each speech unit candidate. Or (2) if the adjacent segment does not have the second speech segment, the phoneme environment parameter of the speech segment candidate corresponding to the segment of interest and the corresponding to the adjacent segment A second cost calculating unit that calculates the connection cost from phoneme environment parameters of a speech unit candidate;
A speech unit selection unit that selects a plurality of speech unit candidates having a low total cost of the target cost and the connection cost as the first speech unit among the plurality of speech unit candidates in the segment of interest. When,
A speech synthesizer. The selection unit includes:
A plurality of the segments as one group, a segment cost calculation unit that calculates a segment cost consisting of the total of the target cost and the connection cost for each segment;
A total cost calculation unit for calculating a total cost consisting of a total of segment costs of each segment;
A speech unit is selected for each segment, and (1) the second speech unit is selected in a segment having the second speech unit, or (2) the second speech unit is selected. In a segment that does not have, a segment selection unit that selects one speech unit that minimizes the total cost from each speech unit candidate;
A sequence calculation unit for obtaining an optimal unit sequence that is a sequence of the speech unit by associating the selected second speech unit or the speech unit with each segment;
Have
The first cost calculation unit and the second cost calculation unit respectively set the target segment and the adjacent segment for each segment on the optimal segment sequence.
The speech synthesizer according to claim 1. The selection unit includes:
The connection cost is calculated using the phoneme environment parameter of the speech unit and the phoneme environment parameter of the speech unit candidate of the adjacent segment on the optimal unit sequence.
The speech synthesizer according to claim 2. The selection unit includes:
A plurality of the first speech segments are selected according to the order given in advance to each segment.
The speech synthesizer according to claim 2. The order is given so that the segment of voiced sound is higher,
The speech synthesizer according to claim 4. The order is given according to the number of speech units included in the speech unit group.
The speech synthesizer according to claim 4. A storage unit storing phoneme environment parameters for each speech unit of the speech unit group and the speech unit group;
A segment setting unit that sets one segment as a target segment from each of a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
An extraction unit that extracts a plurality of speech units having the same characteristics as the phoneme of the segment of interest from the speech unit group as a third speech unit;
A generating unit that generates a fourth speech unit by fusing the plurality of third speech units;
A parameter calculation unit for calculating a phoneme environment parameter of the fourth speech unit;
A target cost representing a distortion amount of the synthesized speech generated using each of the third speech unit and the fourth speech unit of the segment of interest is set as the third speech unit and the fourth speech unit. A third cost calculating unit for calculating each of the phoneme environment parameters and the prosodic information of the target speech;
The third speech unit and the fourth speech unit of adjacent segments that are adjacent segments before or after the segment of interest, and the third speech unit and the fourth speech unit of the segment of interest. Respective connection costs representing distortion amounts generated when connected are the phoneme environment parameters of the third speech unit and the fourth speech unit of the segment of interest, the third speech unit of the adjacent segment, and the A fourth cost calculation unit for calculating from the phoneme environment parameter of the fourth speech unit;
Among the plurality of third speech units and the fourth speech unit in the segment of interest, a plurality of speech units having a low total cost of the target cost and the connection cost are used as the fifth speech unit. A speech segment selector to select;
A synthesis unit that generates synthesized speech by connecting the fifth speech units generated for each of the segments;
A speech synthesizer. A plurality of the segments as one group, a segment cost calculation unit that calculates a segment cost consisting of the total of the target cost and the connection cost for each segment;
A total cost calculation unit for calculating a total cost consisting of a total of segment costs of each segment;
A segment selection unit that selects a speech unit for each segment, and selects one speech unit that minimizes the total cost from the third speech unit and the fourth speech unit. When,
A sequence calculation unit for obtaining an optimal segment sequence that is a sequence of the speech units by associating the speech units with the segments;
Have
The third cost calculation unit and the fourth cost calculation unit respectively set the target segment and the adjacent segment for each segment on the optimal segment sequence.
The speech synthesizer according to claim 7. A storage function for storing a phoneme environment group and a phoneme environment parameter for each speech unit of the speech unit group;
A selection function for selecting a plurality of first speech segments from the speech segment group for each of a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
A generating function for generating one second speech unit by fusing the plurality of first speech units;
A parameter calculation function for calculating a phoneme environment parameter of the second speech unit;
A synthesis function for generating synthesized speech by connecting the second speech units generated for each of the segments;
Have
The selection function is:
A segment setting function for setting, as a target segment, one segment for selecting the first speech segment from the segments;
An extraction function for extracting a plurality of speech units having the same characteristics as the phoneme of the segment of interest from the speech unit group, as speech unit candidates;
A target cost representing a distortion amount of the synthesized speech generated using each speech unit candidate of the segment of interest is calculated from the phoneme environment parameter of each speech unit candidate and the prosodic information of the target speech. A first cost calculating function to
Calculating each connection cost representing the amount of distortion that occurs when a speech unit of an adjacent segment, which is a segment adjacent before or after the target segment, and each speech unit candidate of the target segment are connected; (1) When the adjacent segment is the second speech unit, the connection cost is calculated from the phoneme environment parameter of the second speech unit and the phoneme environment parameter of each speech unit candidate. Or (2) if the adjacent segment does not have the second speech segment, the phoneme environment parameter of the speech segment candidate corresponding to the segment of interest and the corresponding to the adjacent segment A second cost calculating function for calculating the connection cost from phoneme environment parameters of a speech unit candidate;
Speech unit selection function for selecting, as the first speech unit, a plurality of speech unit candidates having a low total cost of the target cost and the connection cost among the plurality of speech unit candidates in the segment of interest. When,
Is a speech synthesis program that implements a computer. The selection function is:
A plurality of the segments as one group, a segment cost calculation function for calculating a segment cost consisting of a sum of the target cost and the connection cost for each segment;
A total cost calculation function for calculating a total cost consisting of a total of segment costs of each segment;
A speech unit is selected for each segment, and (1) the second speech unit is selected in a segment having the second speech unit, or (2) the second speech unit is selected. A segment selection function for selecting one speech unit that minimizes the total cost from the speech unit candidates in a segment that does not have,
A sequence calculation function for obtaining an optimal unit sequence that is a sequence of the speech unit by associating the selected second speech unit or the speech unit with each segment;
Have
The first cost calculation function and the second cost calculation function respectively set the target segment and the adjacent segment for each segment on the optimum segment sequence.
The speech synthesis program according to claim 9. The selection function is:
The connection cost is calculated using the phoneme environment parameter of the speech unit and the phoneme environment parameter of the speech unit candidate of the adjacent segment on the optimal unit sequence.
The speech synthesis program according to claim 10. The selection function is:
A plurality of the first speech segments are selected according to the order given in advance to each segment.
The speech synthesis program according to claim 10. The order is given so that the segment of voiced sound is higher,
The speech synthesis program according to claim 12. The order is given according to the number of speech units included in the speech unit group.
The speech synthesis program according to claim 4. A storage function for storing a phoneme environment group and a phoneme environment parameter for each speech unit of the speech unit group;
A segment setting function for setting one segment as a target segment from among a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
An extraction function for extracting a plurality of speech units having the same characteristics as the phoneme of the segment of interest from the speech unit group as a third speech unit;
A generation function for generating a fourth speech unit by fusing the plurality of third speech units;
A parameter calculation function for calculating a phoneme environment parameter of the fourth speech unit;
A target cost representing a distortion amount of the synthesized speech generated using each of the third speech unit and the fourth speech unit of the segment of interest is set as the third speech unit and the fourth speech unit. A third cost calculating function for calculating from each phoneme environment parameter and the prosody information of the target speech;
The third speech unit and the fourth speech unit of adjacent segments that are adjacent segments before or after the segment of interest, and the third speech unit and the fourth speech unit of the segment of interest. Respective connection costs representing distortion amounts generated when connected are the phoneme environment parameters of the third speech unit and the fourth speech unit of the segment of interest, the third speech unit of the adjacent segment, and the A fourth cost calculating function for calculating from a phoneme environment parameter of the fourth speech unit;
Among the plurality of third speech units and the fourth speech unit in the segment of interest, a plurality of speech units having a low total cost of the target cost and the connection cost are used as the fifth speech unit. A voice segment selection function to select;
A synthesis function for generating synthesized speech by connecting the fifth speech segments generated for each of the segments;
Is a speech synthesis program that implements a computer. A plurality of the segments as one group, a segment cost calculation function for calculating a segment cost consisting of a sum of the target cost and the connection cost for each segment;
A total cost calculation function for calculating a total cost consisting of a total of segment costs of each segment;
A segment selection function for selecting a speech unit for each segment and selecting one speech unit that minimizes the total cost from the third speech unit and the fourth speech unit. When,
A sequence calculation function for obtaining an optimum segment sequence that is a sequence of the speech unit by associating the speech unit with each segment,
Have
The third cost calculation function and the fourth cost calculation function respectively set the target segment and the adjacent segment for each segment on the optimum segment sequence.
The speech synthesis program according to claim 15. A storage step of storing a phoneme group and a phoneme environment parameter for each phoneme of the phoneme group;
A selection step of selecting a plurality of first speech segments from the speech segment group for each of a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
Generating one second speech unit by fusing the plurality of first speech units;
A parameter calculating step of calculating a phoneme environment parameter of the second speech unit;
Generating a synthesized speech by connecting the second speech units generated for each of the segments; and
Have
The selection step includes
A segment setting step for setting, as a segment of interest, one segment for selecting the first speech segment from the segments;
Extracting a plurality of speech units having the same characteristics as the phoneme of the segment of interest from the speech unit group, as speech unit candidates;
A target cost representing a distortion amount of the synthesized speech generated using each speech unit candidate of the segment of interest is calculated from the phoneme environment parameter of each speech unit candidate and the prosodic information of the target speech. A first cost calculating step,
Calculating respective connection costs representing the amount of distortion generated when connecting speech segments of adjacent segments that are adjacent to the target segment before or after the target segment and the speech unit candidates of the target segment; (1) When the adjacent segment is the second speech unit, the connection cost is calculated from the phoneme environment parameter of the second speech unit and the phoneme environment parameter of each speech unit candidate. Or (2) if the adjacent segment does not have the second speech segment, the phoneme environment parameter of the speech segment candidate corresponding to the segment of interest and the corresponding to the adjacent segment A second cost calculating step of calculating the connection cost from a phoneme environment parameter of a speech segment candidate;
A speech unit selection step of selecting, as the first speech unit, a plurality of speech unit candidates having a low total cost of the target cost and the connection cost among the plurality of speech unit candidates in the segment of interest. When,
A speech synthesis method comprising: A storage step of storing a phoneme group and a phoneme environment parameter for each phoneme of the phoneme group;
A segment setting step for setting one segment as a target segment from each of a plurality of segments obtained by dividing a phoneme sequence corresponding to a target speech to be synthesized by a synthesis unit;
An extraction step of extracting a plurality of speech segments having the same characteristics as the phoneme of the segment of interest from the speech segment group as a third speech segment;
Generating a fourth speech unit by fusing the plurality of third speech units;
A parameter calculating step of calculating a phoneme environment parameter of the fourth speech unit;
A target cost representing a distortion amount of the synthesized speech generated using each of the third speech unit and the fourth speech unit of the segment of interest is set as the third speech unit and the fourth speech unit. A third cost calculating step for calculating each of the phoneme environment parameters and the prosodic information of the target speech;
The third speech unit and the fourth speech unit of adjacent segments that are adjacent segments before or after the segment of interest, and the third speech unit and the fourth speech unit of the segment of interest. Respective connection costs representing distortion amounts generated when connected are the phoneme environment parameters of the third speech unit and the fourth speech unit of the segment of interest, the third speech unit of the adjacent segment, and the A fourth cost calculating step for calculating from the phoneme environment parameter of the fourth speech unit;
Among the plurality of third speech units and the fourth speech unit in the segment of interest, a plurality of speech units having a low total cost of the target cost and the connection cost are used as the fifth speech unit. A speech segment selection step to select;
A synthesis step of generating synthesized speech by connecting the fifth speech segments generated for each of the segments;
A speech synthesis method comprising:

Images