JP2011141470A - Phoneme information-creating device, voice synthesis system, voice synthesis method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a phoneme information-creating device capable of reducing a data amount of phoneme information which is a base of voice to be synthesized, while preventing degradation of synthesized voice quality. <P>SOLUTION: A phoneme information-creating device 300 includes: a frame period determination section 301 for determining a frame period for expressing a time interval, based on rhythm relevant information which is information regarding a rhythm of the voice; and a phoneme information creation section 302 which extracts a feature parameter for expressing a feature of a part in a time frame in a phoneme that is a part of a basic voice, and that is a base of voice synthesis processing for synthesizing the voice, on each of a plurality of time frames positioned so that starting positions of contiguous two time frames may be separated by the determined frame period, and which creates phoneme information including a time sequence data of the extracted feature parameter. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a speech synthesis technique, and more particularly, to a segment information generation apparatus that generates segment information used when speech is synthesized.

  2. Description of the Related Art A speech synthesizer that performs speech synthesis processing that analyzes character string information representing a character string and synthesizes speech represented by the character string according to a rule synthesis method is known. Details of the configuration and operation of this type of speech synthesizer are described in, for example, Non-Patent Document 1 to Non-Patent Document 3, and Patent Document 1 and Patent Document 2.

  As shown in FIG. 1, this type of speech synthesizer includes a language processing unit 1, a prosody generation unit 2, a unit information storage unit 12, a unit selection unit 3, and a waveform generation unit 4. Prepare. The unit information storage unit 12 stores unit information including speech unit information representing speech units and attribute information representing attributes of each speech unit. Here, the speech segment is a part of basic speech (speech generated by humans (natural speech)) that is a basis of speech synthesis processing, and is generated by dividing the basic speech into speech synthesis units.

  The speech unit information is information representing the speech waveform of the speech unit or a feature parameter extracted from the speech unit and representing the feature of the speech unit (e.g., a parameter extracted according to a linear prediction method (linear prediction Analysis parameters) or cepstrum coefficients). In many cases, the speech synthesis unit is a phoneme, a syllable, a semi-syllable such as CV (V is a vowel, and C is a consonant), CVC, or VCV. The attribute information includes the environment (phoneme environment) in the basic speech of each speech unit, and prosodic information (basic frequency (pitch frequency), amplitude, duration length, etc.).

  The speech segment information and the attribute information are generated based on, for example, a speech uttered (spoken) by an announcer or a voice actor. The person (speaker) who uttered the voice that is the basis of the speech segment information is also called the original speaker. Note that details of the lengths of speech segments and speech synthesis units are described in Non-Patent Document 1 and Non-Patent Document 3.

  The language processing unit 1 performs analysis such as morphological analysis, syntax analysis, and reading on the input character string information. Then, the language processing unit 1 uses, as a language analysis processing result, information representing a symbol string representing “reading” such as a phoneme symbol and information representing a morpheme part of speech, utilization, accent type, etc. 2 and the unit selection unit 3.

  The prosody generation unit 2 is based on the language analysis processing result output from the language processing unit 1, and the prosody (sound pitch (pitch frequency), sound length (duration length) of the synthesized speech) ) And information related to sound volume (power) and the like, and the prosody information representing the generated prosody is output to the segment selection unit 3 and the waveform generation unit 4 as target prosody information.

  The segment selection unit 3 selects the segment information from the segment information stored in the segment information storage unit 12 based on the language analysis processing result and the target prosodic information as follows, and selects The segment information is output to the waveform generator 4.

  Specifically, the segment selection unit 3 is information representing the characteristics of the synthesized speech based on the input language analysis processing result and the target prosodic information (hereinafter referred to as “target segment environment”). For each speech synthesis unit. The target segment environment includes the corresponding / preceding / following phonemes, the presence or absence of stress, the distance from the accent core, the pitch frequency, power, duration, cepstrum, MFCC (Mel Frequency Cepstral Coefficients), and MFCC. These Δ amounts (change amounts per unit time) and the like.

  Next, the segment selection unit 3 generates segment information representing a speech unit having a phoneme corresponding to (for example, matching) specific information (mainly corresponding phoneme) included in the obtained target segment environment. A plurality of information is acquired from the information storage unit 12. The acquired segment information is a candidate for segment information used for synthesizing speech.

  Then, the segment selection unit 3 calculates a cost that is an index indicating the appropriateness as the segment information used for synthesizing the speech with respect to the acquired segment information. The cost is a value that represents the difference between the target element environment and the attribute information of the candidate element information, and the value becomes smaller as the difference becomes smaller (that is, the higher the degree of similarity between the two). . As the unit information with a lower cost is used, the synthesized speech becomes a speech with a higher degree of naturalness representing the degree of similarity to the speech uttered by a human. Therefore, the segment selection unit 3 selects the segment information with the lowest calculated cost, and outputs the selected segment information to the waveform generation unit 4.

  Based on the target prosody information supplied from the prosody generation unit 2 and the segment information supplied from the segment selection unit 3, the waveform generation unit 4 generates a speech waveform having a prosody that matches or is similar to the target prosody. Generate synthesized speech by connecting the generated speech waveforms.

  In this specification, generating a speech waveform having a prosody that matches or resembles the target prosody based on the segment information is called prosody control in the sense of controlling the prosody of the speech unit. The speech waveform generated from the segment information is also called a segment waveform for the purpose of distinguishing it from other speech waveforms.

  Incidentally, speech units represented by the segment information supplied from the segment selection unit 3 are classified into speech units composed of voiced sounds and speech segments composed of unvoiced sounds. The method used for performing prosody control for voiced sound and the method used for performing prosody control for unvoiced sound are different from each other. The method used to perform prosodic control on unvoiced sound is, for example, the method described in Non-Patent Document 6 or Non-Patent Document 7.

  On the other hand, a method used to perform prosodic control on voiced sound is, for example, a PSOLA (Pitch Synchronous OverLap Add) method described in Non-Patent Document 4. Further, the method used for performing prosodic control on voiced sound may be the method described in Patent Document 3. According to this method, a complex cepstrum representing the spectral envelope of the original speech waveform is obtained, and a segment waveform having a desired pitch frequency is driven by driving a filter representing the complex cepstrum at a time interval corresponding to the desired pitch frequency. Can be generated.

  Next, with reference to FIG. 2, a segment information generating apparatus that generates segment information stored in the segment information storage unit 12 will be described. As shown in FIG. 2, the segment information generation device includes a recorded sentence storage unit 21, a natural speech storage unit 22, an attribute information extraction unit 7, a feature parameter extraction unit 6, a segment information storage unit 12, .

  The natural speech storage unit 22 stores information representing basic speech (natural speech waveform) that is a basis for generating segment information. The recorded sentence storage unit 21 stores language information including information representing a character string corresponding to the basic speech stored in the natural speech storage unit 22. The language information is, for example, information representing a kanji-kana mixed sentence. Furthermore, the language information may include information such as readings, accent positions, morpheme parts of speech, etc. (for example, information having the same format as the information output by the language processing unit 1).

  The feature parameter extraction unit 6 extracts time series data of feature parameters from the natural speech waveform supplied from the natural speech storage unit 22, and outputs the extracted time series data to the segment information storage unit 12. The extraction method depends on a method (waveform generation method) used by the waveform generation unit 4 of the speech synthesizer shown in FIG. 1 to generate a speech waveform. For example, when the waveform generation unit 4 uses the PSOLA method described in Non-Patent Document 4 as the waveform generation method, the feature parameter extraction unit 6 uses the method described in Non-Patent Document 4 as the feature parameter extraction method. It is preferable to use it.

  In addition, when a linear prediction analysis parameter or a cepstrum coefficient is used as the feature parameter, the waveform generation unit 4 uses the method described in Non-Patent Document 5. The attribute information extraction unit 7 acquires attribute information based on the natural speech waveform supplied from the natural speech storage unit 22 and the linguistic information supplied from the recorded sentence storage unit 21, and acquires the acquired attribute information. The information is output to the single information storage unit 12.

  When the speech unit information generated by the unit information generating device is not time-series data of feature parameters but information representing a speech waveform, each feature parameter is a window as described in Non-Patent Document 5. Generated by applying a function to a natural speech waveform.

  Specifically, the feature parameter extraction unit 6 extracts feature parameters representing features of a portion in a time frame having a predetermined time length (frame length) in the natural speech waveform. The feature parameter extraction unit 6 acquires feature parameters extracted for each of a plurality of time frames as time series data of feature parameters. The plurality of time frames are arranged such that the start positions of two consecutive time frames are separated by a predetermined frame period.

  Here, the frame length and the frame period value differ depending on the method used to extract the feature parameter. When the value synchronized with the pitch frequency of the natural speech waveform from which the feature parameter is extracted is used as the value of the frame length and frame period (pitch frequency synchronization method), it is always without synchronizing with the pitch frequency. There are cases where a fixed value is used (pitch frequency asynchronous method). An example of the pitch frequency synchronization method is the PSOLA method.

  On the other hand, when extracting a linear prediction analysis parameter or a cepstrum coefficient, a pitch frequency asynchronous method is used. For example, the segment information generating apparatus described in Patent Document 4 extracts feature parameters using a pitch frequency asynchronous method. This segment information generation device uses empirically determined values for the frame period and the frame length.

JP 2005-91551 A JP 2006-84854 A Japanese Patent No. 2812184 JP 2003-66982 A

Huang, Acero, Hon, “Spoken Language Processing”, Prentice Hall, 2001, p.689-836 Ishikawa, “Basics of Prosodic Control for Speech Synthesis”, IEICE Technical Report, IEICE, 2000, 100, 392, p.27-34 Abe, “Basics of Synthesis Units for Speech Synthesis”, IEICE Technical Report, IEICE, 2000, 100, 392, p.35-42 Moulines, Charapentier, “Pitch-Synchronous Waveform Processing Techniques For Text-To-Speech Synthesis Using Diphones”, Speech Communication, 1990, Vol. 9, p.435-467 Furui, “Speech Information Processing”, Morikita Publishing, 1998 R. Suzuki, M. Misaki, “Time-scale modification of speech signals using cross-correlation functions”, IEEE Trans. Consum. Electron., IEEE, 1992, Vol. 38, p.357-363 Kiyoyama et al., “Development of a high-quality real-time speech rate conversion system”, IEICE Transactions, 2001, J84-D-II, No. 6, p.918-926

  However, when speech unit information (here, feature parameter time-series data) is generated using an empirically determined frame period, it is synthesized when speech is synthesized using the speech unit information. Audio quality (sound quality) may be low. For example, if the frame period is set to an excessive value, the frequency with which feature parameters are extracted decreases. As a result, the continuity in the time direction of the characteristic parameters used when performing the speech synthesis process becomes insufficient (that is, the characteristic parameters do not change sufficiently smoothly with respect to time), so that the sound quality is deteriorated. This phenomenon will be described with reference to FIG.

  FIG. 3 is an explanatory diagram conceptually showing a process of synthesizing speech by extracting time series data of feature parameters from natural speech and reconstructing the extracted time series data at a desired pitch period. . The left column in FIG. 3 represents the case where the frame period is relatively short, and the right column represents the case where the frame period is relatively long.

  As shown in FIG. 3, as the frame period becomes shorter, the number of feature parameters (amount of time-series data) extracted from a natural speech waveform having a certain length of time increases. In this example, feature parameters are extracted for each of the seven time frames when the frame period is short, and feature parameters are extracted for each of the three time frames when the frame period is long.

  When synthesizing speech by reconstructing time-series data of feature parameters at a desired pitch period, the speech synthesizer selects a feature parameter for each pitch period and generates a speech waveform from the selected feature parameter. The pitch synchronization position shown in FIG. 3 is a position (time) arranged such that the interval between other adjacent pitch synchronization positions coincides with the pitch period.

  In this example, when the frame period is short (the left column in FIG. 3), the speech synthesizer selects each of the time frames having frame numbers 2, 3,. On the other hand, if the frame period is long (the right column in FIG. 3), the speech synthesizer selects two time frames with a frame number of 1, selects three time frames with a frame number of 2, and the frame number is Select one of the three time frames. The speech synthesizer synthesizes speech based on speech unit information extracted for the selected time frame.

  In this example, the speech synthesizer selects the feature parameter closest to the pitch synchronization position when selecting the feature parameter for each pitch period. That is, when the frame period is short (the left column in FIG. 3), the speech synthesizer ensures the continuity of the feature parameters in the time direction without repeatedly selecting the same time frame (that is, the feature parameter). (Ie, the feature parameters change smoothly enough over time). On the other hand, when the frame period is long (the right column in FIG. 3), the speech synthesizer repeatedly selects the same time frame. As a result, the continuity of the feature parameter in the time direction is impaired (that is, the feature parameter does not change sufficiently smoothly with respect to time).

  That is, when the frame period is excessively long, the number of feature parameters is insufficient when synthesizing speech. In other words, in this case, the speech synthesizer synthesizes speech based on an excessively small amount of information. Therefore, in order to ensure sufficient sound quality, it is necessary to shorten the frame period sufficiently. However, when the frame period is excessively shortened, the data amount of the piece information is excessively increased. As a result, the storage capacity required for the storage device that stores the piece information becomes excessively large.

  As described above, according to the segment information generation device described in the above-described document, there is a problem that the quality of synthesized speech is deteriorated or a data amount of segment information is excessively increased. It was.

  For this reason, the object of the present invention is to solve the above-mentioned problem “when the quality of the synthesized speech is reduced or the data amount of the piece information is excessively increased”. An object of the present invention is to provide a segment information generation apparatus.

In order to achieve such an object, an element information generating apparatus according to one aspect of the present invention is
Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process that synthesizes speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
Is provided.

  With the configuration as described above, the present invention can reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being degraded. .

It is a figure showing the schematic structure of the speech synthesizer which concerns on background art. It is a figure showing the schematic structure of the segment information generation apparatus which concerns on background art. It is explanatory drawing which showed notionally the process which synthesize | combines audio | voice by extracting the time series data of the characteristic parameter from natural audio | voice, and reconstructing the extracted time series data with a desired pitch period. 1 is a diagram illustrating a schematic configuration of a segment information generation device according to a first embodiment of the present invention. FIG. 5 is a diagram illustrating a schematic configuration of a segment information generation device according to a second embodiment of the present invention. FIG. 10 is a diagram illustrating a schematic configuration of a segment information generating apparatus according to a third embodiment of the present invention. FIG. 10 is a diagram illustrating a schematic configuration of a segment information generation device according to a fourth embodiment of the present invention. FIG. 10 is a diagram illustrating a schematic configuration of a speech synthesizer according to a fifth embodiment of the present invention. It is a figure showing the schematic structure of the segment information generation apparatus which concerns on 6th Embodiment of this invention.

  Hereinafter, each embodiment of a segment information generation device, a speech synthesis system, a speech synthesis method, and a program according to the present invention will be described with reference to FIGS.

<First embodiment>
(Constitution)
As shown in FIG. 4, the segment information generating apparatus 100 according to the first embodiment of the present invention includes a recorded sentence storage unit 21, a natural speech storage unit 22, an attribute information extraction unit 7, and a feature parameter extraction unit. 6, segment information storage unit 12, frame cycle storage unit 13, frame cycle determination unit (frame cycle determination unit) 31, synthesized speech prosody information estimation unit (prosody information estimation unit) 33, synthesized speech prosody information And a storage unit 34. The feature parameter extraction unit 6 and the attribute information extraction unit 7 constitute unit information generation means.

  The natural speech storage unit 22 stores information representing basic speech (natural speech waveform) that is a basis for generating segment information. The unit information includes speech unit information representing a speech unit and attribute information representing an attribute of each speech unit. Here, a speech segment is a part of basic speech (speech generated by humans (natural speech)) that is the basis of speech synthesis processing that synthesizes speech, and is divided by speech synthesis units. Generated.

  In this example, the speech unit information includes time-series data of feature parameters extracted from the speech unit and representing the features of the speech unit. The speech synthesis unit is a syllable. Note that the speech synthesis unit may be a phoneme, a semi-syllable such as CV (V is a vowel, C is a consonant), CVC, or VCV. The attribute information includes the environment (phoneme environment) in the basic speech of each speech unit, and prosodic information (basic frequency (pitch frequency), amplitude, duration length, etc.).

  The recorded sentence storage unit 21 stores language information including information representing a character string (recorded sentence) corresponding to the basic speech stored in the natural speech storage unit 22. The language information is, for example, information representing a kanji-kana mixed sentence. Furthermore, the language information may include information such as readings, accent positions, morpheme parts of speech.

  The feature parameter extraction unit 6 extracts time series data of feature parameters from the natural speech waveform supplied from the natural speech storage unit 22. Specifically, the feature parameter extraction unit 6 extracts feature parameters representing features of a portion in a time frame having a preset time length (frame length) in the natural speech waveform.

  The feature parameter extraction unit 6 acquires feature parameters extracted for each of a plurality of time frames as time series data of feature parameters. In the plurality of time frames, the start positions of two consecutive time frames are separated by a frame period determined by the frame period determination unit 31 (a frame period stored in the frame period storage unit 13 as described later). Placed in. Here, the frame period represents a time interval. The feature parameter extraction unit 6 outputs the acquired time series data to the unit information storage unit 12 as speech unit information.

The attribute information extraction unit 7 acquires attribute information based on the natural speech waveform supplied from the natural speech storage unit 22 and the linguistic information supplied from the recorded sentence storage unit 21, and acquires the acquired attribute information. The information is output to the single information storage unit 12.
That is, it can be said that the feature parameter extraction unit 6 and the attribute information extraction unit 7 generate segment information including time series data of feature parameters.

  The segment information storage unit 12 stores segment information including the speech segment information supplied from the feature parameter extraction unit 6 and the attribute information supplied from the attribute information extraction unit 7.

  The synthesized speech prosody information storage unit 34 is a speech synthesizer (not shown) that performs speech synthesis processing based on the unit information stored in the unit information storage unit 12 (that is, generated by the unit information generation device 100). The prosodic information output from the speech synthesis processing means) is stored in advance. That is, the synthesized speech prosody information storage unit 34 stores in advance prosody information representing the prosody of past synthesized speech that is speech synthesized by the speech synthesizer at a past time.

  Based on the statistics of the prosodic information stored in the synthesized speech prosody information storage unit 34, the synthesized speech prosody information estimation unit 33 (to be described later) is a speech that is synthesized by the speech synthesizer at a future time. Prosody information representing the prosody of is estimated. Accordingly, it is preferable that the information amount of the prosodic information stored in the synthesized speech prosodic information storage unit 34 is a sufficiently large amount.

  The synthesized speech prosody information estimation unit 33 estimates the prosody information representing the prosody of the future synthesized speech based on the prosody information representing the prosody of the past synthesized speech stored in the synthesized speech prosody information storage unit. The synthesized speech prosody information estimation unit 33 outputs the estimated prosody information to the frame period determination unit 31 as prosody related information. The prosody related information is information related to the prosody of the speech.

  By the way, it is not possible to specify text (character string) to be subjected to speech synthesis processing performed at a future time. Therefore, in this example, the synthesized speech prosody information estimation unit 33 estimates prosody information representing the speech prosody synthesized by the speech synthesizer when an arbitrary text is input at a future time point.

  Note that the prosody of the speech synthesized by the speech synthesizer when any text is input has a value within a certain range. Therefore, it is considered preferable to estimate prosodic information representing the prosody of the future synthesized speech based on the statistics of the prosodic information representing the prosody of the past synthesized speech.

  Therefore, the synthesized speech prosody information estimation unit 33 calculates a statistic of prosodic information representing the prosody of the past synthesized speech. The synthesized speech prosody information estimation unit 33 estimates the calculated statistics of the prosody information representing the prosody of the past synthesized speech as the prosodic information representing the prosody of the future synthesized speech.

  Here, the statistic of the prosodic information is an average value, variance, frequency (histogram), maximum value, minimum value, mode value, or the like of the pitch frequency or the duration length. These statistics may be calculated uniquely from the entire prosodic information stored in the synthesized speech prosodic information storage unit 34, or calculated for each unit such as phonemes, syllables, semi-syllables, or segments. May be.

  In this example, the synthesized speech prosody information estimation unit 33 calculates an average value and variance of pitch frequencies for each speech synthesis unit as a statistic of prosodic information.

  The frame period determining unit 31 determines a frame period for each speech synthesis unit (that is, a preset processing unit) based on the prosody related information supplied from the synthesized speech prosody information estimation unit 33. The frame cycle determination unit 31 outputs the determined frame cycle to the frame cycle storage unit 13.

  By the way, the higher the pitch frequency of the speech synthesized by speech synthesis processing (synthesized speech), the greater the number of time frames (number of frames) required to sufficiently increase the quality (sound quality) of the synthesized speech. . That is, it is preferable to reduce the frame period as the average value of the pitch frequency increases.

In addition, the greater the variance of the pitch frequency for a certain speech synthesis unit, the higher the possibility that the sound quality will deteriorate due to the insufficient number of frames.
Therefore, in this example, the frame period determination unit 31 determines the frame period based on the average value and variance of the pitch frequencies.

Specifically, the frame period determination unit 31 determines the frame period for each speech synthesis unit according to Equation 1. Here, F0 _mean is an average value of pitch frequencies represented by prosody related information, and F0 _var is a variance of pitch frequencies represented by prosody related information. T _p is a frame period, and β ₁ is a preset threshold value. α ₁ is a constant having a preset positive value (ie, greater than 0). T _p1 is a first frame period having a preset positive value, and T _p2 is a second frame period having a preset positive value. Furthermore, the first frame period T _p1 is smaller than the second frame period T _p2 . That is, the first frame period T _p1 and the second frame period T _p2 satisfy 0 <T _p1 <T _p2 .

  Thus, the frame period determination unit 31 determines a value that decreases as the pitch frequency represented by the prosodic information increases as the frame period. Specifically, the frame period determining unit 31 determines, as the frame period, a value that decreases as the average value of the pitch frequency represented by the prosodic related information increases and decreases as the pitch frequency dispersion represented by the prosodic related information increases. To do.

  In this example, the frame period determination unit 31 is configured to select one frame period from two frame periods, but is configured to select one frame period from three or more frame periods. May be. In this case, the frame period can be controlled with higher accuracy. As a result, it is possible to further reduce the data amount of the segment information that is the basis of the synthesized speech while reliably preventing the quality of the synthesized speech from being lowered.

  In this example, the frame period determination unit 31 is configured to determine the frame period based on the average value and variance of the pitch frequency, but determines the frame period based only on the average value of the pitch frequency. It may be configured as follows.

  The frame cycle storage unit 13 stores the frame cycle (value) output from the frame cycle determination unit 31 in association with the segment information. As described above, the frame period stored in the frame period storage unit 13 is a value used when the feature parameter extraction unit 6 extracts feature parameters, and the speech synthesizer executes speech synthesis processing. It is a value used in the case.

(Operation)
Next, the operation of the segment information generating apparatus 100 configured as described above will be described.
First, the speech synthesizer generates prosodic information by performing speech synthesis processing, and outputs the generated prosodic information to the segment information generating device 100. The synthesized speech prosody information storage unit 34 stores the prosody information output from the speech synthesizer. By repeating such processing, the synthesized speech prosody information storage unit 34 accumulates a sufficiently large amount of prosodic information.

  Thereafter, the user of the segment information generating apparatus 100 inputs information instructing to generate segment information. Thereby, the segment information generating apparatus 100 starts a process of generating segment information.

  First, the synthesized speech prosody information estimation unit 33 calculates the average pitch frequency for each speech synthesis unit as prosody information representing the prosody of the future synthesized speech based on the prosody information stored in the synthesized speech prosody information storage unit 34. Calculate (estimate) values and variances.

Next, the synthesized speech prosody information estimation unit 33 outputs the estimated prosody information to the frame period determination unit 31 as prosody related information.
Then, the frame period determination unit 31 determines a frame period for each speech synthesis unit based on the prosody related information output from the synthesized speech prosody information estimation unit 33. Next, the frame period storage unit 13 stores the frame period determined by the frame period determination unit 31.

  Then, the feature parameter extraction unit 6 is based on the natural speech waveform stored in the natural speech storage unit 22, and the start positions of two consecutive time frames are only the frame period stored in the frame cycle storage unit 13. A feature parameter for each of a plurality of time frames arranged to be separated from each other is extracted. Further, the attribute information extraction unit 7 acquires attribute information.

  In this way, the feature parameter extraction unit 6 and the attribute information extraction unit 7 generate unit information composed of speech unit information including time series data of feature parameters and attribute information. Then, the segment information storage unit 12 stores the segment information generated by the feature parameter extraction unit 6 and the attribute information extraction unit 7.

  After that, when receiving the text to be subjected to the speech synthesis process, the speech synthesizer synthesizes speech for synthesizing the speech representing the received text based on the segment information stored in the segment information storage unit 12. Process.

  As described above, according to the segment information generating apparatus 100 according to the first embodiment, the frame period used when generating segment information can be appropriately set based on the prosody of speech. . As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

Further, in the first embodiment, the synthesized speech prosody information estimation unit 33 performs the prosody of the future synthesized speech based on the prosody information representing the prosody of the past synthesized speech that is the speech synthesized at the past time point by the speech synthesis processing means. The prosodic information representing is estimated.
According to this, it is possible to appropriately determine the frame period based on a prosody that is relatively likely to be used in speech synthesis processing at a future time.

In addition, in the first embodiment, the frame cycle determination unit 31 determines a value that decreases as the pitch frequency (fundamental frequency) increases as the frame cycle.
Thereby, it is possible to reliably prevent the quality of the synthesized voice from being deteriorated.

Furthermore, in the first embodiment, the frame period determination unit 31 acquires prosodic related information for each processing unit set in advance and determines a frame period based on the acquired prosodic related information.
According to this, for example, the frame period can be set for each syllable or for each semi-syllable. As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

<Second Embodiment>
Next, the segment information generating apparatus according to the second embodiment of the present invention will be described. The segment information generation device according to the second embodiment is configured to provide prosody information representing the prosody of the future synthesized speech based on the prosody information representing the prosody of the basic speech, with respect to the segment information generation device according to the first embodiment. There is a difference in estimation. Accordingly, the following description will focus on such differences.

  As shown in FIG. 5, the segment information generation device 100 according to the second embodiment includes an appearance number calculation unit (appearance number acquisition unit) in addition to the configuration of the segment information generation device 100 according to the first embodiment. ) 46. Also, the segment information generating apparatus 100 according to the second embodiment includes a frame period determining unit 41 that replaces the frame period determining unit 31, a synthetic speech prosody information estimating unit 43 that replaces the synthesized speech prosody information estimating unit 33, and a synthesized speech A segment prosody information storage unit 44 instead of the prosody information storage unit 34;

The attribute information extraction unit 7 acquires attribute information for each speech unit, and outputs prosodic information (segment prosody information) included in the acquired attribute information to the segment prosody information storage unit 44.
The segment prosody information storage unit 44 stores the segment prosody information supplied from the attribute information extraction unit 7 for each speech unit.

  The number-of-appearance calculation unit 46 is the number of times the voice corresponding to the voice synthesis unit appears in the basic voice for each voice synthesis unit (processing unit) based on the language information stored in the recorded sentence storage unit 21. Calculate (acquire) the number of occurrences. The appearance number calculation unit 46 calculates the number of appearances for each speech synthesis unit by counting the number of times a character string corresponding to the speech synthesis unit for which the appearance number is calculated appears in the recorded sentence.

For example, when the number of occurrences of “wa” is calculated in the case where the speech synthesis unit is a syllable, the appearance number calculation unit 46 determines that “wa” is included in the recorded sentence represented by the language information stored in the recorded sentence storage unit 21. Count the number of occurrences.
The appearance number calculation unit 46 outputs the calculated number of appearances to the synthesized speech prosody information estimation unit 43.

  The synthesized speech prosody information estimation unit 43 is supplied from the prosody information for each speech unit stored in the segment prosody information storage unit 44 (that is, the prosody information representing the prosody of the basic speech) and the appearance number calculation unit 46. The prosody information representing the prosody of the future synthesized speech for each speech unit is estimated based on the number of appearances for each speech synthesis unit. The synthesized speech prosody information estimation unit 43 outputs the estimated prosody information to the frame period determination unit 41 as prosody related information.

  By the way, the speech synthesizer is configured to select segment information having a prosody closest to the target prosody that is the target value of the synthesized speech. Therefore, there is little difference between the prosody of a certain speech unit in the synthesized speech and the prosody of the speech unit in the basic speech that is the basis of the speech unit in the future.

  However, for speech synthesis units with a relatively small number of appearances, the speech synthesizer cannot select segment information having a prosody sufficiently close to the target prosody. Therefore, the prosody of a certain speech unit in the synthesized speech in the future may be relatively different from the prosody of the speech unit in the basic speech that is the basis of the speech unit.

  In other words, the degree to which the prosody of a certain speech unit in the synthesized speech and the prosody of the speech unit in the basic speech on which the speech unit is based varies depending on the number of appearances. I can say that. Therefore, the synthesized speech prosody information estimation unit 43 estimates prosody information representing the prosody of the future synthesized speech for each speech unit using such properties.

In this example, the synthesized speech prosody information estimation unit 43 estimates prosodic information representing the prosody of the future synthesized speech for each speech unit according to Equation 2. In this example, a case where the prosody to be estimated is a pitch frequency will be described.

Here, N (u) is the number of appearances, which is the number of times the speech unit s corresponding to the speech synthesis unit u appears in the basic speech. F0 _est (s) is an estimated value of the prosody (pitch frequency) of the future synthesized speech for the speech unit s, and F0 _org (s) is the prosody of the basic speech for the speech unit s. Β ₂ is a preset threshold value. F0 _mod1 is a first coefficient having a preset positive value, and F0 _mod2 is a second coefficient having a preset positive value. Furthermore, the first coefficient F0 _mod1 is smaller than the second coefficient F0 _mod2 . That is, the first coefficient F0 _mod1 and the second coefficient F0 _mod2 satisfy 0 <F0 _mod1 <F0 _mod2 .

In this way, the synthesized speech prosody information estimation unit 43 multiplies the prosody information representing the basic speech prosody by the first coefficient F0 _mod1 or the second coefficient F0 _mod2, and the prosody representing the future synthesized speech prosody. Estimate as information. In addition, the synthesized speech prosody information estimation unit 43 estimates a value that decreases as the number of appearances increases as prosody information representing the prosody of the synthesized speech in the future.

  The frame cycle determination unit 41 determines the frame cycle based on the prosodic information estimated by the synthesized speech prosody information estimation unit 43, similarly to the frame cycle determination unit 31 according to the first embodiment.

  As described above, in the segment information generating apparatus 100 according to the second embodiment, the synthesized speech prosody information estimation unit 43 uses the prosodic information representing the prosody of the future synthesized speech based on the prosodic information representing the prosody of the basic speech. Is estimated.

  By the way, it is relatively likely that the segment information used in the speech synthesis process is information having a prosody similar (as close as possible) to the prosody of the synthesized speech. Therefore, by configuring the segment information generating apparatus 100 as described above, prosody information representing the prosody of a synthesized speech in the future can be estimated with high accuracy. As a result, the frame period can be appropriately determined based on the prosody that is relatively likely to be used in speech synthesis processing at a future time.

  Furthermore, in the second embodiment, the synthesized speech prosody information estimation unit 43 represents the prosody of the future synthesized speech based on the appearance number acquired by the appearance number calculation unit 46 and the prosody information representing the prosody of the basic speech. Estimate prosodic information.

  By the way, as the number of appearances increases for a certain speech synthesis unit, segment information representing a speech unit having a higher prosody that matches the prosody of the synthesized speech is used in the speech synthesis process. . Therefore, by configuring the segment information generating apparatus 100 as described above, prosodic information representing the prosody of a synthesized speech in the future can be estimated with higher accuracy. As a result, the frame period can be appropriately determined based on the prosody that is more likely to be used in the speech synthesis process at a future time.

  In the second embodiment, the segment information generation device 100 is configured to estimate prosody information representing the prosody of a future synthesized speech based on the number of appearances calculated for each speech synthesis unit. The prosody information representing the prosody of the synthesized speech in the future may be estimated based on the statistics (average value and / or variance) of the number of appearances calculated for each unit. In this case, for example, the synthesized speech prosody information estimation unit 43 replaces the number of appearances with a value based on the statistics of the number of appearances (for example, the product of the variance of the number of appearances and a preset coefficient) It is preferable to use a value added to the value).

  In the second embodiment, the unit information generation apparatus 100 is configured to estimate prosody information representing the prosody of the synthesized speech based on prosody information representing the prosody of the basic speech. The prosody information representing the prosody of the synthesized speech in the future may be estimated based on the statistics of the prosody information representing the prosody.

For example, the statistic of prosodic information is an average value, variance, frequency (histogram), maximum value, minimum value, mode value, or the like of pitch frequency or duration time.
Here, a description will be given of the segment information generating apparatus 100 according to the modified example configured to estimate the prosodic information representing the prosody of the synthesized speech based on the average value and the variance of the prosodic information representing the prosody of the basic speech. .

The synthesized speech prosody information estimation unit 43 according to this modification estimates prosody information representing the future synthesized speech prosody according to Equation 3. Here, F0 _est (u) is an estimated value of the prosody (pitch frequency in this example) of the future synthesized speech for the speech synthesis unit u. F0 _m (u) is the average value of the prosody of the speech unit corresponding to the speech synthesis unit u in the basic speech, and F0 _v (u) is the speech synthesis unit u in the basic speech. Is the variance of the prosody of the speech segment corresponding to. Α ₂ is a constant having a preset positive value.

  Even the segment information generating apparatus 100 configured as described above can achieve the same operations and effects as those of the segment information generating apparatus 100 according to the second embodiment.

<Third embodiment>
Next, the segment information generating apparatus according to the third embodiment of the present invention will be described. The segment information generation device according to the second embodiment is different from the segment information generation device according to the second embodiment in that the frame period is determined without estimating the prosody of the synthesized speech in the future. . Accordingly, the following description will focus on such differences.

  As shown in FIG. 6, the segment information generation apparatus 100 according to the third embodiment does not include the synthesized speech prosody information estimation unit 43 included in the segment information generation apparatus 100 according to the second embodiment. Furthermore, the segment information generating apparatus 100 according to the third embodiment includes a frame period determining unit 51 instead of the frame period determining unit 41.

  The frame period determining unit 51 acquires prosodic information for each speech unit (that is, prosodic information representing the prosody of the basic speech) stored in the segment prosodic information storage unit 44 as prosodic related information. Further, the frame period determination unit 51 determines the frame period based on the acquired prosodic information and the number of appearances for each speech synthesis unit supplied from the appearance number calculation unit 46. The frame cycle determination unit 51 outputs the determined frame cycle to the frame cycle storage unit 13.

Also in this example, a case where a pitch frequency is used as prosodic information will be described.
In this example, the frame period determination unit 51 determines the frame period for each speech unit according to Equation 4. Here, F0 _org (s) is the basic speech prosody for the speech unit s corresponding to the speech synthesis unit u. Β ₃ (s) is a threshold value for the speech segment s. T _p3 is a third frame period having a preset positive value, and T _p4 is a fourth frame period having a preset positive value. Furthermore, the third frame period T _p3 is smaller than the fourth frame period T _p4 . That is, the third frame period T _p3 and the fourth frame period T _p4 satisfy 0 <T _p3 <T _p4 .

  As described above, the frame period determination unit 51 determines a value that decreases as the pitch frequency represented by the prosodic information increases as the frame period. In this example, the frame period determining unit 51 is configured to select one frame period from two frame periods, but is configured to select one frame period from three or more frame periods. May be. In this case, the frame period can be controlled with higher accuracy. As a result, it is possible to further reduce the data amount of the segment information that is the basis of the synthesized speech while reliably preventing the quality of the synthesized speech from being lowered.

Further, the frame period determination unit 51 determines the threshold β ₃ (s) for the speech unit s based on the appearance number N (u) that is the number of times the speech unit s corresponding to the speech synthesis unit u appears in the basic speech. To decide. In this example, the frame period determination unit 51 determines the threshold value β ₃ (s) for the speech element s according to Equation 5.

Here, β ₀ is a constant having a preset positive value, and β ₄ is a preset threshold value. Β _mod1 is a first coefficient having a preset positive value, and β _mod2 is a second coefficient having a preset positive value. Furthermore, the first coefficient β _mod1 is smaller than the second coefficient β _mod2 . That is, the first coefficient β _mod1 and the second coefficient β _mod2 satisfy 0 <β _mod1 <β _mod2 .

As described above, the frame period determination unit 51 determines a value that decreases as the appearance number N (u) increases as the threshold value β ₃ (s). Therefore, the frame period determining unit 51 determines a value that decreases as the pitch frequency represented by the prosodic information increases and decreases as the appearance number decreases as the frame period.

  As described above, in the segment information generation device 100 according to the third embodiment, the frame period determination unit 51 acquires prosody information representing the prosody of the basic speech as prosody related information, and the acquired prosody related information. To determine the frame period.

  By the way, it is relatively likely that the segment information used in the speech synthesis process is information having a prosody similar (as close as possible) to the prosody of the synthesized speech. Therefore, by configuring the segment information generating apparatus 100 as described above, the frame period can be appropriately determined based on the prosody that is relatively likely to be used in the speech synthesis process.

  Furthermore, in the third embodiment, the frame cycle determination unit 51 determines the frame cycle based on the acquired number of appearances and prosodic information.

  By the way, with respect to a certain processing unit (speech synthesis unit), as the number of appearances increases, the unit information representing a speech unit having a higher prosody that matches the prosody of the synthesized speech becomes speech synthesis. Used in processing. Therefore, by configuring the segment information generating apparatus 100 as described above, the frame period can be appropriately determined based on the prosody that is more likely to be used in the speech synthesis process.

  In the third embodiment, the frame period determination unit 51 is configured to acquire prosody information representing the prosody of the basic speech as prosody related information, but the prosody information representing the prosody of the past synthesized speech is You may be comprised so that it may acquire as information.

  By the way, the prosody of a voice synthesized at a future time is relatively likely to be similar to the prosody of a voice synthesized at a past time. Therefore, by configuring the segment information generating apparatus 100 as described above, it is possible to appropriately determine the frame period based on a prosody that is relatively likely to be used in speech synthesis processing at a future time. .

<Fourth embodiment>
Next, an element information generating apparatus according to the fourth embodiment of the present invention will be described. The segment information generating apparatus according to the fourth embodiment is different from the segment information generating apparatus according to the first embodiment in that the synthesized speech prosody that is the prosody of the past synthesized speech and the basic speech prosody that is the prosody of the basic speech. Is different in that the frame period is determined based on the prosodic control amount representing the difference. Accordingly, the following description will focus on such differences.

  As shown in FIG. 7, the segment information generation apparatus 100 according to the fourth embodiment does not include the synthesized speech prosody information estimation unit 33 included in the segment information generation apparatus 100 according to the first embodiment. Furthermore, the segment information generation device 100 according to the fourth embodiment includes a frame period determination unit 61 that replaces the frame period determination unit 31, and a prosody control amount information storage unit 64 that replaces the synthesized speech prosody information storage unit 34. .

  The prosody control amount information storage unit 64 is a speech synthesizer (not shown) that performs speech synthesis processing based on the unit information stored in the unit information storage unit 12 (that is, generated by the unit information generation device 100). The prosody control amount output from the speech synthesis processing means) is stored in advance.

  The prosody control amount is information indicating a difference between a synthesized speech prosody that is a prosody of speech synthesized in the past by the speech synthesizer (past synthesized speech) and a basic speech prosody that is a prosody of the basic speech. is there. In this example, the prosody control amount information storage unit 64 stores the magnitude of the difference between the basic speech prosody and the synthesized speech prosody as a prosody control amount.

  Specifically, when a pitch frequency is used as a prosody, when a basic speech pitch frequency is 250 Hz and a past synthesized speech pitch frequency is 300 Hz, a prosodic control amount information storage is performed. The unit 64 stores 50 (= | 300−250 |) as the prosody control amount.

  When the duration is used as a prosody, when a duration of the basic speech is 250 ms and a duration of the past synthesized speech is 300 ms for a certain speech unit, the prosodic control amount information storage is performed. The unit 64 stores 50 (= | 300−250 |) as the prosody control amount. The prosody control amount information storage unit 64 uses a value based on the ratio between the basic speech prosody and the synthesized speech prosody (a value that increases as the difference between the basic speech prosody and the synthesized speech prosody increases) as the prosody control amount. It may be configured to store.

  The frame period determination unit 61 calculates (acquires) a statistic of the prosodic control amount based on the prosodic control amount stored in the prosody control amount information storage unit 64. Here, the statistic of the prosodic control amount is an average value, variance, frequency (histogram), maximum value, minimum value, or mode value of the prosodic control amount. These statistics may be calculated uniquely from the entire prosody control amount stored in the prosody control amount information storage unit 64, and for each unit such as phoneme, syllable, semi-syllable, or segment. It may be calculated.

  In this example, the frame period determination unit 61 calculates the average value and variance of the prosodic control amount for each speech synthesis unit as the prosodic control amount statistics. In this example, the frame period determination unit 61 uses the magnitude of the difference between the duration of the past synthesized speech and the duration of the basic speech as the prosody control amount.

  The frame period determining unit 61 acquires the calculated average value and variance of the prosodic control amount as prosodic related information. The frame period determination unit 61 determines a frame period based on the acquired prosodic information, and outputs the determined frame period to the frame period storage unit 13.

  By the way, the higher the prosody control amount, the lower the quality of synthesized speech. Therefore, the frame period determining unit 61 determines a value that decreases as the prosodic control amount increases as the frame period. As a result, it is possible to reliably prevent the quality of the synthesized voice from deteriorating.

  In this example, the frame cycle determination unit 61 determines a value that becomes smaller as the value obtained by multiplying the variance of the prosodic control amount by a preset coefficient to the average value of the prosodic control amount increases as the frame cycle. . The frame period determining unit 61 may be configured to determine the frame period based only on the average value of the prosodic control amount.

  As described above, in the segment information generating apparatus 100 according to the fourth embodiment, the frame period determining unit 61 acquires the prosody control amount representing the difference between the synthesized speech prosody and the basic speech prosody as the prosody related information. The frame period is determined based on the acquired prosodic information. As a result, the frame period can be reduced as the prosodic control amount increases. As a result, it is possible to reliably prevent the quality of the synthesized voice from deteriorating.

<Fifth embodiment>
Next, a speech synthesis system according to a fifth embodiment of the present invention will be described. The speech synthesis system according to the fifth embodiment is a system that performs speech synthesis processing based on the segment information generated by the segment information generation device according to the first embodiment. Hereinafter, the configuration other than the unit information generation apparatus will be mainly described.

  This speech synthesis system includes the segment information generation device 100 according to the first embodiment and the speech synthesis device 200 shown in FIG. The speech synthesizer 200 includes a language processing unit 1, a prosody generation unit 2, a segment selection unit 3, and a waveform generation unit 8.

  The language processing unit 1 performs analysis such as morphological analysis, syntax analysis, and reading on the input character string information. Then, the language processing unit 1 uses, as a language analysis processing result, information representing a symbol string representing “reading” such as a phoneme symbol and information representing a morpheme part of speech, utilization, accent type, etc. 2 and the unit selection unit 3.

  The prosody generation unit 2 is based on the language analysis processing result output from the language processing unit 1, and the prosody (sound pitch (pitch frequency), sound length (duration length) of the synthesized speech) ) And information related to sound volume (power) and the like, and the prosody information representing the generated prosody is output to the segment selection unit 3 and the waveform generation unit 8 as target prosody information.

  The segment selection unit 3 selects the segment information from the segment information stored in the segment information storage unit 12 based on the language analysis processing result and the target prosodic information as follows, and selects The segment information is output to the waveform generator 8.

  Specifically, the segment selection unit 3 is information representing the characteristics of the synthesized speech based on the input language analysis processing result and the target prosodic information (hereinafter referred to as “target segment environment”). For each speech synthesis unit. The target segment environment includes the corresponding / preceding / following phonemes, the presence / absence of stress, the distance from the accent core, the pitch frequency for each speech synthesis unit, power, duration, cepstrum, MFCC (Mel Frequency Cepstral Coefficients), and These Δ amounts (change amounts per unit time) and the like.

  Next, the segment selection unit 3 generates segment information representing a speech unit having a phoneme corresponding to (for example, matching) specific information (mainly corresponding phoneme) included in the obtained target segment environment. A plurality of information is acquired from the information storage unit 12. The acquired segment information is a candidate for segment information used for synthesizing speech.

  Then, the segment selection unit 3 calculates a cost that is an index indicating the appropriateness as the segment information used for synthesizing the speech with respect to the acquired segment information. The cost is a value that represents the difference between the target element environment and the attribute information of the candidate element information, and the value becomes smaller as the difference becomes smaller (that is, the higher the degree of similarity between the two). . As the unit information with a lower cost is used, the synthesized speech becomes a speech with a higher degree of naturalness representing the degree of similarity to the speech uttered by a human. Therefore, the segment selection unit 3 selects the segment information with the lowest calculated cost, and outputs the selected segment information to the waveform generation unit 8.

  The waveform generation unit 8 is based on the target prosody information supplied from the prosody generation unit 2, the segment information supplied from the unit selection unit 3, and the frame period stored in the frame period storage unit 13. Then, a speech waveform having a prosody that matches or is similar to the target prosody is generated, and the generated speech waveform is connected to generate a synthesized speech.

  In this way, the speech synthesizer 200 synthesizes speech based on the frame period determined by the segment information generator 100 and the segment information generated by the segment information generator 100. The speech synthesis processing means for performing is configured.

  As described above, according to the speech synthesis system according to the fifth embodiment, the frame period used when generating the segment information can be appropriately set based on the prosody of the speech. As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

<Sixth embodiment>
Next, an element information generating apparatus according to a sixth embodiment of the present invention will be described with reference to FIG.
The segment information generating apparatus 300 according to the sixth embodiment is
A frame period determining unit (frame period determining means) 301 for determining a frame period representing a time interval based on prosody related information which is information related to the prosody of a speech;
A part of the basic speech that is the basis of the speech synthesis process that synthesizes speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. A segment information generation unit (segment information) that extracts a feature parameter representing a feature of a portion of a certain speech unit within the time frame and generates unit information including time-series data of the extracted feature parameter. Generation means) 302,
Is provided.

  According to this, the frame period used when generating the segment information can be appropriately set based on the prosody of the speech. As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

  Although the present invention has been described with reference to the above embodiment, the present invention is not limited to the above-described embodiment. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  In each of the above embodiments, each function of the segment information generation apparatus is realized by hardware such as a circuit. By the way, the segment information generation device includes a processing device and a storage device that stores a program (software), and the processing device is configured to implement each function by executing the program. Also good. In this case, the program may be stored in a computer-readable recording medium. For example, the recording medium is a portable medium such as a flexible disk, an optical disk, a magneto-optical disk, and a semiconductor memory.

  In addition, as another modified example of the above-described embodiment, any combination of the above-described embodiments and modified examples may be employed.

  The present invention is applicable to a speech synthesis system that performs speech synthesis processing for synthesizing speech.

<Appendix>
(Appendix 1)
Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
A unit information generating apparatus comprising:

  According to this, the frame period used when generating the segment information can be appropriately set based on the prosody of the speech. As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

(Appendix 2)
An element information generation device according to appendix 1,
Prosody information estimating means for estimating prosody information representing the prosody of a future synthesized speech that is a speech synthesized at a future time by speech synthesis processing means for performing the speech synthesis processing based on the generated segment information. ,
The segment information generation device configured to acquire the estimated prosodic information as the prosodic information and determine the frame period based on the acquired prosodic related information.

(Appendix 3)
An element information generation device according to appendix 2,
The prosodic information estimation means estimates prosodic information representing the prosody of the future synthesized speech based on the prosodic information representing the prosody of the past synthesized speech that is the speech synthesized at the past time by the speech synthesis processing means. The unit information generation apparatus comprised in this.

  By the way, the prosody of a voice synthesized at a future time is relatively likely to be similar to the prosody of a voice synthesized at a past time. Therefore, by configuring the segment information generation apparatus as described above, the frame period can be appropriately determined based on a prosody that is relatively likely to be used in speech synthesis processing at a future time.

(Appendix 4)
An element information generation device according to appendix 2,
The segment information generating device configured to estimate the prosodic information representing the prosody of the future synthesized speech based on the prosodic information representing the prosody of the basic speech.

  By the way, it is relatively likely that the segment information used in the speech synthesis process is information having a prosody similar (as close as possible) to the prosody of the synthesized speech. Therefore, by configuring the segment information generation apparatus as described above, prosody information representing the prosody of a synthesized speech in the future can be estimated with high accuracy. As a result, the frame period can be appropriately determined based on the prosody that is relatively likely to be used in speech synthesis processing at a future time.

(Appendix 5)
An element information generation device according to appendix 4,
For each processing unit set in advance, an appearance number acquisition means for acquiring an appearance number that is the number of times the sound corresponding to the processing unit appears in the basic sound,
The prosody information estimating means is configured to estimate prosody information representing the prosody of the future synthesized speech based on the acquired number of appearances and the prosody information representing the prosody of the basic speech. Generator.

  By the way, for a certain processing unit (for example, syllables, phonemes, semi-syllables such as CV (V is a vowel, C is a consonant), CVC, or VCV, etc.) The unit information representing the speech unit having a higher prosody with the same degree is used in the speech synthesis process.

  Therefore, by configuring the segment information generation apparatus as described above, prosody information representing the prosody of a synthesized speech in the future can be estimated with higher accuracy. As a result, the frame period can be appropriately determined based on the prosody that is more likely to be used in the speech synthesis process at a future time.

(Appendix 6)
An element information generation device according to appendix 1,
The frame period determining means converts prosodic information representing the prosody of past synthesized speech that is speech synthesized at a past time point by speech synthesis processing means for performing the speech synthesis processing based on the segment information. A segment information generation device configured to determine the frame period based on the acquired prosodic information.

  By the way, the prosody of a voice synthesized at a future time is relatively likely to be similar to the prosody of a voice synthesized at a past time. Therefore, by configuring the segment information generation apparatus as described above, the frame period can be appropriately determined based on a prosody that is relatively likely to be used in speech synthesis processing at a future time.

(Appendix 7)
An element information generation device according to appendix 1,
The frame period determining unit is configured to acquire prosody information representing the prosody of the basic speech as the prosody related information and determine the frame period based on the acquired prosody related information. .

  By the way, it is relatively likely that the segment information used in the speech synthesis process is information having a prosody similar (as close as possible) to the prosody of the synthesized speech. Therefore, by configuring the segment information generation apparatus as described above, the frame period can be appropriately determined based on a prosody that is relatively likely to be used in speech synthesis processing.

(Appendix 8)
An element information generation device according to appendix 7,
For each processing unit set in advance, an appearance number acquisition means for acquiring an appearance number that is the number of times the sound corresponding to the processing unit appears in the basic sound,
The segment information generation device configured to determine the frame period based on the acquired number of appearances and the prosodic information, the frame period determining unit.

  By the way, for a certain processing unit (for example, syllables, phonemes, semi-syllables such as CV (V is a vowel, C is a consonant), CVC, or VCV, etc.) The unit information representing the speech unit having a higher prosody with the same degree is used in the speech synthesis process. Therefore, by configuring the segment information generating apparatus as described above, the frame period can be appropriately determined based on the prosody that is more likely to be used in the speech synthesis process.

(Appendix 9)
The element information generation device according to any one of appendix 2 to appendix 8,
The segment information generating device configured to acquire the statistic of the prosodic information as the prosodic related information.

(Appendix 10)
The element information generation device according to any one of appendix 2 to appendix 9,
The prosody information includes information representing a fundamental frequency,
The unit information generating apparatus configured to determine, as the frame period, a value that decreases as the fundamental frequency increases.

  By the way, in order to sufficiently improve the quality of synthesized speech, it is necessary to shorten the frame period as the fundamental frequency (pitch frequency) increases. Therefore, by configuring the segment information generation apparatus as described above, it is possible to reliably prevent the quality of synthesized speech from being deteriorated.

(Appendix 11)
An element information generation device according to appendix 1,
The frame period determining means is a synthesized speech prosody that is a speech prosody synthesized at a past time by a speech synthesis processing means that performs the speech synthesis processing based on the segment information, and a prosody of the basic speech A segment information generation device configured to acquire a prosody control amount representing a difference from a basic speech prosody as the prosody related information, and to determine the frame period based on the acquired prosodic related information.

  By the way, the higher the prosody control amount, the lower the quality of synthesized speech. Therefore, by configuring the segment information generation apparatus as described above, the frame period can be reduced as the prosodic control amount increases. As a result, it is possible to reliably prevent the quality of the synthesized voice from deteriorating.

(Appendix 12)
An element information generation device according to appendix 11,
The prosody control amount is an amount that increases as a difference between the synthesized speech prosody and the basic speech prosody increases.
The unit information generating apparatus configured to determine, as the frame period, a value that decreases as the prosodic control amount increases.

(Appendix 13)
The element information generation device according to appendix 11 or appendix 12,
The segment information generation device configured to acquire the statistic of the prosodic control amount as the prosodic related information.

(Appendix 14)
The element information generation device according to any one of appendix 1 to appendix 13,
The unit information generating apparatus configured to acquire the prosodic information for each preset processing unit and to determine the frame period based on the acquired prosodic related information.

  According to this, for example, the frame period can be set for each syllable or for each semi-syllable. As a result, it is possible to reduce the data amount of the segment information that is the basis of the synthesized voice while preventing the quality of the synthesized voice from being lowered.

(Appendix 15)
Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
Speech synthesis processing means for performing speech synthesis processing for synthesizing speech based on the determined frame period and the generated segment information;
A speech synthesis system comprising:

(Appendix 16)
The speech synthesis system according to appendix 15,
Prosody information estimation means for estimating prosody information representing the prosody of a future synthesized voice that is a voice synthesized at a future time by the voice synthesis processing means,
The speech synthesis system configured to acquire the estimated prosodic information as the prosodic information and determine the frame period based on the acquired prosodic related information.

(Appendix 17)
The speech synthesis system according to appendix 15,
The frame period determining means acquires prosodic information representing the prosody of past synthesized speech that is speech synthesized at a past time by the speech synthesis processing means as the prosodic related information, and based on the acquired prosodic related information A speech synthesis system configured to determine the frame period.

(Appendix 18)
The speech synthesis system according to appendix 15,
The speech synthesis system configured to acquire the prosody information representing the prosody of the basic speech as the prosody related information, and to determine the frame cycle based on the acquired prosody related information.

(Appendix 19)
The speech synthesis system according to appendix 15,
The frame period determining means is a prosodic control amount representing a difference between a synthesized speech prosody that is a speech prosody synthesized at a past time by the speech synthesis processing means and a basic speech prosody that is a prosody of the basic speech. Is obtained as the prosodic related information, and the frame period is determined based on the acquired prosodic related information.

(Appendix 20)
A frame period representing a time interval is determined based on prosodic information, which is information related to speech prosody,
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. A speech synthesis method for extracting feature parameters representing features of a portion of a certain speech unit within the time frame and generating unit information including time-series data of the extracted feature parameters.

(Appendix 21)
The speech synthesis method according to appendix 20, wherein
Estimating prosodic information representing the prosody of a future synthesized speech that is speech synthesized at a future time by speech synthesis processing means for performing the speech synthesis processing based on the generated segment information;
A speech synthesis method configured to acquire the estimated prosodic information as the prosodic related information and determine the frame period based on the acquired prosodic related information.

(Appendix 22)
The speech synthesis method according to appendix 20, wherein
Prosody information representing the prosody of past synthesized speech that is speech synthesized at a past time point by speech synthesis processing means that performs the speech synthesis processing based on the unit information is acquired as the prosody related information, and the acquired A speech synthesis method configured to determine the frame period based on prosodic information.

(Appendix 23)
The speech synthesis method according to appendix 20, wherein
A speech synthesis method configured to acquire prosody information representing the prosody of the basic speech as the prosody related information and to determine the frame period based on the acquired prosody related information.

(Appendix 24)
The speech synthesis method according to appendix 20, wherein
A difference between a synthesized speech prosody that is a speech prosody synthesized at a past time by speech synthesis processing means that performs the speech synthesis processing based on the unit information, and a basic speech prosody that is a prosody of the basic speech A speech synthesis method configured to acquire a prosodic control amount representing the prosodic information as the prosodic information and determine the frame period based on the acquired prosodic information.

(Appendix 25)
In the information processing device,
Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
A program to realize

(Appendix 26)
The program according to attachment 25, wherein
In addition to the information processing apparatus,
Realization of prosodic information estimation means for estimating prosodic information representing the prosody of a future synthesized speech that is a speech synthesized at a future time by speech synthesis processing means for performing the speech synthesis processing based on the generated segment information As well as
The program configured to acquire the estimated prosodic information as the prosodic information and determine the frame period based on the acquired prosodic related information.

(Appendix 27)
The program according to attachment 25, wherein
The frame period determining means converts prosodic information representing the prosody of past synthesized speech that is speech synthesized at a past time point by speech synthesis processing means for performing the speech synthesis processing based on the segment information. And a program configured to determine the frame period based on the acquired prosodic information.

(Appendix 28)
The program according to attachment 25, wherein
The frame period determining means is a program configured to acquire prosodic information representing the prosody of the basic speech as the prosodic related information and determine the frame period based on the acquired prosodic related information.

(Appendix 29)
The program according to attachment 25, wherein
The frame period determining means is a synthesized speech prosody that is a speech prosody synthesized at a past time by a speech synthesis processing means that performs the speech synthesis processing based on the segment information, and a prosody of the basic speech A program configured to acquire a prosody control amount representing a difference from a basic speech prosody as the prosody related information and determine the frame period based on the acquired prosodic related information.

1 Language processor
2 Prosody generator
3 Element selector
4 Waveform generator
6 Feature parameter extraction unit
7 Attribute information extractor
8 Waveform generator
12 Segment information storage
13 Frame cycle memory
21 Recorded sentence storage
22 Natural voice memory
31 Frame period determination unit
33 Synthetic speech prosody information estimation part
34 Synthetic speech prosody information storage
41 Frame period determination unit
43 Synthetic speech prosody information estimation part
44 fragment prosody information storage
46 Appearance count calculator
51 Frame period determination unit
61 Frame period determination unit
64 Prosodic control amount information storage
100 fragment information generator
200 speech synthesizer
300 Segment information generator
301 Frame period determination unit
302 Segment information generator

Claims (10)

Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
A unit information generating apparatus comprising: The element information generation device according to claim 1,
Prosody information estimating means for estimating prosody information representing the prosody of a future synthesized speech that is a speech synthesized at a future time by speech synthesis processing means for performing the speech synthesis processing based on the generated segment information. ,
The segment information generation device configured to acquire the estimated prosodic information as the prosodic information and determine the frame period based on the acquired prosodic related information. The element information generation device according to claim 2,
The prosodic information estimation means estimates prosodic information representing the prosody of the future synthesized speech based on the prosodic information representing the prosody of the past synthesized speech that is the speech synthesized at the past time by the speech synthesis processing means. The unit information generation apparatus comprised in this. The element information generation device according to claim 2,
The segment information generating device configured to estimate the prosodic information representing the prosody of the future synthesized speech based on the prosodic information representing the prosody of the basic speech. The element information generation device according to claim 1,
The frame period determining means converts prosodic information representing the prosody of past synthesized speech that is speech synthesized at a past time point by speech synthesis processing means for performing the speech synthesis processing based on the segment information. A segment information generation device configured to determine the frame period based on the acquired prosodic information. The element information generation device according to claim 1,
The frame period determining unit is configured to acquire prosody information representing the prosody of the basic speech as the prosody related information and determine the frame period based on the acquired prosody related information. . The element information generation device according to claim 1,
The frame period determining means is a synthesized speech prosody that is a speech prosody synthesized at a past time by a speech synthesis processing means that performs the speech synthesis processing based on the segment information, and a prosody of the basic speech A segment information generation device configured to acquire a prosody control amount representing a difference from a basic speech prosody as the prosody related information, and to determine the frame period based on the acquired prosodic related information. Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
Speech synthesis processing means for performing speech synthesis processing for synthesizing speech based on the determined frame period and the generated segment information;
A speech synthesis system comprising: A frame period representing a time interval is determined based on prosodic information, which is information related to speech prosody,
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. A speech synthesis method for extracting feature parameters representing features of a portion of a certain speech unit within the time frame and generating unit information including time-series data of the extracted feature parameters. In the information processing device,
Frame period determining means for determining a frame period representing a time interval based on prosody related information which is information relating to the prosody of speech;
A part of the basic speech that is the basis of the speech synthesis process for synthesizing speech for each of a plurality of time frames arranged so that the start positions of two consecutive time frames are separated by the determined frame period. Unit information generating means for extracting a feature parameter representing a feature of a portion of a certain speech unit in the time frame and generating unit information including time-series data of the extracted feature parameter;
A program to realize

Images