JP2010224419A - Voice synthesizer, method and, program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To further reproduce and synthesizes features of an original voice in an environment where an available data size or communication bit rates are extremely limited. <P>SOLUTION: This voice synthesizer synthesizes a voice based on a series of information for voice synthesis. The information for voice synthesis is composed of common information for synthesizing a voice regardless of target vocal quality and correction information which functions when performing the voice synthesis of predetermined vocal quality. The voice synthesizer includes: a correction means which modifies contents of the common information based on the correction information when synthesizing a voice with predetermined vocal quality; and a voice synthesis means which synthesizes a voice waveform based on the corrected information; and further includes: a control command generation means which generates a control command for the generation of the voice waveform based on the common information; a control command correction means which corrects the generated control command based on the correction information when synthesizing the voice with predetermined vocal quality; and a voice waveform generation means which generates the voice waveform based on the corrected control command. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a speech synthesizer, a method, and a program for generating a speech waveform based on features of speech recorded in advance.

A CELP (Code Excited Linear Prediction) method is known as a highly efficient speech coding method specialized for speech. The CELP method is based on the knowledge about the physical characteristics of the speech waveform, but does not directly use speech language restrictions, so it can efficiently encode any style of speech in any language. Has possible characteristics. However, the bit rate of encoded audio is at least several kbps (bits
per second). On the other hand, the technology for synthesizing speech from linguistic information generally belongs to speech synthesis technology. Text-to-speech is a typical method of using speech synthesis technology. Here, for example, a symbol representing phoneme types and prosodic features obtained by analyzing text is used as the input. A device for generating a speech waveform is particularly called a speech synthesizer, and symbols constituting its input are called speech synthesis symbols. There are various forms of symbols for speech synthesis. Here, let us consider a case in which phonological information constituting a series of speech and prosodic information mainly expressed as a pose or voice pitch are simultaneously described. As an example of such a speech synthesis symbol, there is JEITA (Electronic Information Technology Industry Association) standard IT-4002 “symbol for Japanese text speech synthesis” (Non-patent Document 1).

  There are various methods of generating a speech waveform in the speech synthesizer. Here, two typical methods are described: a connection synthesis method and an analysis synthesis method.

  In the connection synthesis method, a large amount of speech is collected in advance, and its speech fragments (hereinafter referred to as speech segments) are stored in a database in advance. In this method, a speech unit having a good connection relationship with a segment is selected from the segment database and synthesized. Each speech segment is assigned parameters such as phoneme information, acoustic parameters, and appearance environment in the speech corpus.

  In the connection synthesis method, the speech unit to be used is selected (hereinafter referred to as segment selection) based on synthesis target information specified by the speech synthesis symbol string. This unit selection is referred to as cost. This is performed based on a distortion measure, that is, an index indicating the degree of deterioration of the speech waveform synthesized by the selected speech segment from the target synthesized speech waveform. The cost can usually be divided into a target cost indicating an error between the synthesis target information and the speech unit, and a connection cost indicating the degree of discontinuity between the speech units, and segment selection minimizes the overall cost. To be done.

  As for the unit database, the speech waveform may be stored as it is, or data obtained by compressing the unit by CELP or the like may be stored. When the speech waveform is accumulated, the final output speech waveform is synthesized by connecting the segments on the waveform. On the other hand, when a segment is compressed using CELP or the like, not only the segment is connected on the waveform obtained by decoding the segment, but also connected at the parameter level before decoding and decoded together. There is also a technique for generating a speech waveform.

  On the other hand, the analysis and synthesis method is a method in which an operation is performed on a feature parameter of speech obtained as a result of analyzing speech, and a speech waveform is generated from the feature parameter by signal processing. Here, in particular, a method of synthesizing a speech waveform by signal processing based on a sound source / filter model in which a sound source and a filter are combined, similar to the CELP method, is targeted. In the sound source / filter model, the sound waveform is synthesized in a signal processing manner by driving a filter that generates sound of sound with an appropriate sound source, but here, unlike the CELP method, relatively, such as an impulse train or a white noise source. Considering the case of driving with a simple sound source. Hereinafter, the sound source parameters and the filter parameters are collectively referred to as speech synthesis parameters. The speech synthesis parameter is composed of a plurality of parameters such as a MFCC (Mel-Frequency Cepstral Coefficient) for expressing spectral features and a fundamental frequency (F0) of the waveform corresponding to the voice pitch. As the filter, an AR (autoregressive) type filter, an MLSA (Mel logarithmic spectrum approximation) filter (Non-Patent Document 2), which directly uses MFCC as a parameter, and the like are used.

  For example, in order to synthesize speech such as consonants, it is necessary to change the speech synthesis parameters over time. In this method, for example, the speech synthesis parameters are updated at a fixed period of about 5 ms and the characteristics are changed. It is common to synthesize speech while One period of this fixed period is generally called one frame. Therefore, in order to synthesize speech with this configuration, it is necessary to determine the value of the speech synthesis parameter for each frame from the speech synthesis symbol.

  As one of the methods, a time change of a speech synthesis parameter time series is modeled based on an appropriate model, and the model parameter is generated by first predicting from a speech synthesis symbol. A method is used in which an arbitrary speech can be synthesized by generating a sequence. Hereinafter, this model is referred to as a speech generation model. For example, the speech synthesis parameter feature of a phoneme is divided into three states in terms of time. For the number of frames in each state, the vectors representing the statistical features are d1, d2, and d3 in order from the first state. Speech synthesis parameters for synthesizing the phoneme if the vector elements are concatenated to create one vector d, and the vectors representing the statistical features of each state are v1, v2, and v3 in order from the first state. Can be expressed by four vectors d, v1, v2, and v3 that constitute the parameters of the speech generation model.

  As described above, it is assumed that all phonemes can be represented by four vectors in this way, and an optimal codebook is created for each vector in advance. Alternatively, v1, v2, and v3 may be expressed using the same codebook. At the time of speech synthesis, first, an optimal codebook vector constituting the parameters of the speech generation model of each phoneme is predicted from the speech synthesis symbols, and a speech generation model is constructed to synthesize each phoneme. Then, these speech generation models are connected in time order to obtain a speech generation model for one utterance, and an optimal speech synthesis parameter time series is obtained based on the model. A voice waveform is generated by controlling the sound source / filter based on the voice synthesis parameter time series.

  As described above, by using the speech synthesis method as described above, a speech waveform can be generated from data of about several hundred bps expressed in the form of a speech synthesis symbol string.

"Symbols for Japanese text-to-speech synthesis" JEITA standard IT-4002, March 2005 Sei Imai, Kazuo Sumita, Chieko Furuichi, "Mel Log Spectrum Approximation (MLSA) Filter for Speech Synthesis", IEICE Transactions (A), J66-A, 2, pp.122-129, Feb. 1983 .

  Consider a case where a voice message prepared in advance is reproduced in an environment where the available data size and communication bit rate are very limited. In particular, here, the case where audio data is stored in a target having a limited storage capacity, such as a two-dimensional barcode, is mainly targeted. The preference for voice quality and reading style is considered to vary from user to user, so it is desirable to prepare various voice quality and reading styles as variations in advance in order to increase user satisfaction. . In the following, for the sake of explanation, it is considered that all differences between individual variations are “voice quality” differences. Therefore, “voice quality” in the following is not limited in the sense of phonetic definition, but represents a more generalized voice personality.

  If there is no limitation on the data size, the speech may be encoded independently for each voice quality by the conventional speech encoding technique. However, in the conventional speech coding method, a certain amount of data size is required to maintain the quality, and speech is coded for each voice quality difference, so the data size is proportional to the number of voice qualities. For this reason, it is difficult to increase the number of voice qualities while ensuring quality with a limited data size.

  On the other hand, by using a speech synthesis technique, it is possible to synthesize speech using only a speech synthesis symbol string, although the size of the speech synthesis apparatus is somewhat large. A voice synthesizer that outputs voice of a certain voice quality can be configured by collecting a large amount of voice data of the same kind of voice quality in advance on a scale of, for example, several tens of minutes to several tens of hours. A speech synthesis symbol string is usually much smaller than speech-encoded data and can be easily recorded on a two-dimensional barcode or the like. Furthermore, by constructing data for speech synthesizers from multiple voice quality data in advance and incorporating them into the device, the voice quality of the output speech can be compared by switching the data when using the device. Can be changed easily. However, in general, a speech synthesizer synthesizes speech by predicting the features of the target speech from only linguistic information constituting the speech synthesis symbol string, and compared with the case where natural speech is speech-encoded, There is a problem that the reproducibility of the voice becomes low and the naturalness is greatly inferior.

  As described above, in the related art, when the storage capacity is limited, it is impossible to achieve both the naturalness of the voice message and the diversity of the voice quality.

  Therefore, the present invention provides a speech synthesizer, method, and program capable of reproducing and synthesizing the features of the original speech in an environment where the available data size and communication bit rate are very limited. Objective.

In order to achieve the above object, a speech synthesizer according to the present invention is a speech synthesizer that synthesizes speech based on a series of speech synthesis information, and the speech synthesis information synthesizes speech regardless of the target voice quality. The correction information functioning when synthesizing a predetermined voice quality, and when synthesizing a voice of a predetermined voice quality, the correction means for changing the content of the common information based on the correction information, Speech synthesis means for synthesizing a speech waveform based on the corrected information;
Is provided.

  In order to achieve the above object, a speech synthesizer according to the present invention is a speech synthesizer that synthesizes speech based on a series of speech synthesis information, and the speech synthesis information synthesizes speech regardless of the target voice quality. Control information generating means configured to generate a control command for generating a speech waveform based on the common information, and voice of a predetermined voice quality. In the case of synthesizing, a control command correcting unit that corrects the generated control command based on the correction information, and a voice waveform generating unit that generates a voice waveform based on the corrected control command.

  The common information is preferably composed of phonological symbols and prosodic symbols.

  The speech synthesizer preferably further includes means for selecting the voice quality of the synthesized voice from a plurality of voice qualities.

  In order to achieve the above object, a speech synthesis method according to the present invention is a speech synthesis method for synthesizing speech based on a series of speech synthesis information, and the speech synthesis information is used to synthesize speech regardless of target voice quality. The correction information that functions when synthesizing a predetermined voice quality, and when synthesizing a voice of a predetermined voice quality, a correction step of changing the content of the common information based on the correction information; and A speech synthesis step of synthesizing a speech waveform based on the corrected information.

  In order to achieve the above object, a speech synthesis method according to the present invention is a speech synthesis method for synthesizing speech based on a series of speech synthesis information, and the speech synthesis information is used to synthesize speech regardless of target voice quality. Control information generating step for generating a control command for generating a speech waveform based on the common information, and a voice of a predetermined voice quality. When synthesizing, a control command correcting step of correcting the generated control command based on the correction information and a voice waveform generating step of generating a voice waveform based on the corrected control command are provided.

  In order to achieve the above object, a program according to the present invention causes a computer to function as the speech synthesizer described above.

  According to the present invention, by dividing the information for speech synthesis into common information that is information independent of voice quality and correction information that differs for each voice quality, only the correction information is shared by sharing the common information for the voice of the same utterance content. Can be efficiently encoded. At this time, by generating correction information that makes the command information for generating the voice waveform approach the characteristics of the original natural voice, a highly natural voice that reproduces the characteristics of the original voice is output. be able to.

  Further, when the common information is composed only of linguistic information similar to the conventional speech synthesis technology, even if correction information corresponding to the voice quality desired by the user is not included in the speech synthesis information, Since speech synthesis is possible, the naturalness is reduced, but speech of the voice quality desired by the user can be synthesized.

  Moreover, even if the linguistic expression is different in a part of the utterance, the common information can be shared by setting the common information to be corrected.

  Based on the above features, it is possible to reduce the size of the required storage capacity compared to the conventional speech coding method and to produce a plurality of speech with different speech qualities with similar utterance content than when using the conventional speech synthesis technology as it is. It is possible to reproduce and synthesize voice characteristics.

1 is a block diagram of a speech synthesizer according to a first embodiment of the present invention. 2 shows a flowchart of speech synthesis according to the first embodiment. It is a block diagram of the speech synthesizer by the 2nd Embodiment of this invention. 10 shows a flowchart of speech synthesis according to the second embodiment. It is a block diagram of the speech synthesizer by the 3rd Embodiment of this invention. 10 shows a flowchart of speech synthesis according to a third embodiment.

  The best mode for carrying out the present invention will be described in detail below with reference to the drawings. In the following, the “unit speech” is a minimum speech processing unit in the speech synthesizer according to the present invention. Specific examples of unit speech include phonemes, syllables, and words. However, here, unit speech is classified in consideration of differences in phonemic environment such as the types of phonemes before and after, and differences in prosodic features such as accent, intonation, and speech speed. A “speech synthesis symbol” is a series of symbols for describing each type of unit speech included in one speech.

  FIG. 1 is a block diagram of a speech synthesizer according to the first embodiment of the present invention. As shown in FIG. 1, the speech synthesizer 1 includes a preprocessing unit 11 for speech synthesis information, a common information correction unit 12, and a speech waveform synthesis unit 13. FIG. 2 shows a flowchart of speech synthesis according to the first embodiment.

S11. First, the preprocessing unit 11 processes the information for speech synthesis, and extracts common information and correction information. As an example of the common information, there is a symbol for simultaneously describing phonetic information for speech synthesis and prosodic information, such as a symbol defined in JEITA IT-4002. Although speech recognition technology may be used to create common information, generally speaking, since speech is often read out from a prepared document, Japanese text-to-speech conversion technology is used for such common information. Thus, it can be automatically created with relatively high accuracy from a voice manuscript mixed with kanji characters. Also, errors caused by automatic processing and errors that occur between actual utterances can be easily corrected manually. The correction information includes identification information for identifying the target voice quality.
S12. Here, the common information correction unit 12 receives the common information and the correction information that are output from the preprocessing unit 11 as inputs. The common information correction unit 12 confirms whether the waveform generation target voice quality of the voice waveform synthesis unit 13 matches the target voice quality of the correction information.
S13. If they match, the common information correction unit 12 corrects and outputs the common information based on the correction information. On the other hand, when the waveform generation target voice quality of the voice waveform synthesis unit 13 is different from the target voice quality of the correction information, the common information correction unit 12 outputs the input common information as it is.
S14. Based on the output of the common information correction unit 12, the speech waveform synthesis unit 13 synthesizes a speech waveform and outputs it as the output of the speech synthesizer 1.

  Examples of correction information include a code indicating that the i-th symbol from the top of the common information is replaced with the symbol A, a code indicating that the symbol B is inserted immediately before the j-th symbol from the top, and k from the top. There is a list of codes that delete the th symbol. At the time of correction, the common information is corrected by performing the corresponding operations in order from the top code. For example, when common information such as syllable is used as a unit, “Densha.” (Where the symbol “'” has an accent nucleus at that place is a head-high accent in the common information. ) Is read with a flat accent in the voice information subject to the correction information, the head information accent “Densha.” Can be obtained by setting one correction code to delete the second symbol. The flat accent “Densha.” Is corrected. By the same correction, it is possible to correct the difference in the pose insertion position and the small difference in expression such as “De'Sne.” And “De'Swa.”.

  FIG. 3 is a block diagram of a speech synthesizer according to the second embodiment of the present invention. As shown in FIG. 3, the speech synthesizer 1 includes a preprocessing unit 11, a speech synthesis control command generation unit 14, a control command correction unit 15, and a speech waveform generation unit 16. The combination of the speech synthesis control command generation unit 14 and the speech waveform generation unit 16 here corresponds to the speech waveform synthesis unit 13 in the first embodiment. FIG. 4 shows a flowchart of speech synthesis according to the second embodiment.

S21. The pre-processing unit 11 processes the information for speech synthesis and extracts common information and correction information.
S22. The voice synthesis control command generation unit 14 converts the extracted common information into a voice synthesis control command for voice waveform generation.
S23. When the control command is input, the control command correction unit 15 checks whether the waveform generation target voice quality of the voice waveform generation unit 16 matches the target voice quality of the correction information.
S24. If they match, the control command correction unit 15 corrects the control command based on the separately input correction information. If they do not match, the control command correction unit 15 outputs the control command as it is.
S25. A speech waveform is generated by the speech waveform generation unit 16 based on the output information of the control command correction unit 15 and output as an output of the speech synthesizer 1.

  Here, as an example of the control command, there is a segment ID when the connection synthesis method is used as the voice synthesis method. In this case, the segment selection process required in the connection synthesis method is performed by the speech synthesis control command generation unit 14, and the process of connecting the segments is performed by the speech waveform generation unit 16. On the other hand, when the analysis / synthesis method is used, a speech synthesis parameter time series is exemplified as the control command. Specifically, this is a parameter time series of MFCC and F0. In this case, the speech synthesis control command generation unit 14 performs processing for generating a speech synthesis parameter time series from the speech generation model.

  Alternatively, the parameters of the voice generation model may be used as the control command here. In this case, the speech waveform generation unit 16 performs processing for generating a speech synthesis parameter time series from the speech generation model.

  In addition, as an example of the correction information in this case, there is a string of codes representing a replacement operation regarding a certain value of each parameter. Also in this embodiment, the correction information includes identification information for identifying the target voice quality.

  By configuring the correction information so that the characteristics related to the spectrum of the voice output from the speech synthesizer 1 and the characteristics related to the fundamental frequency approach each characteristic of the original voice, compared to the case where no correction is performed, It is possible to output sound closer to the characteristics of the original sound.

  For example, correction of values is attempted for all elements of all feature vectors, and as a result, the correction of the trial in which the synthesized speech feature is closest to the original speech feature is corrected by 1 One element. (1) The number of corrections reaches a predetermined number of times, or the size of the entire correction information reaches a predetermined limit value, or (2) The difference in characteristics between the synthesized voice and the original voice is predetermined. The correction information can be created by repeatedly accumulating the elements of the correction information until it becomes smaller than the standard.

  Alternatively, a device configuration in which the above two correction processes are combined may be used, and a voice synthesis control command in which both pieces of correction information are mixed may be configured. FIG. 5 is a block diagram of a speech synthesizer according to the third embodiment of the present invention having this configuration. As shown in FIG. 5, the speech synthesizer 1 includes a preprocessing unit 11, a common information correction unit 12, a voice synthesis control command generation unit 14, a control command correction unit 15, and a voice waveform generation unit 16. FIG. 6 shows a flowchart of speech synthesis according to the third embodiment.

S31. The pre-processing unit 11 processes the information for speech synthesis and extracts common information and correction information.
S32. The common information correction unit 12 confirms whether the waveform generation target voice quality of the voice waveform generation unit 16 matches the target voice quality of the correction information.
S33. If they match, the common information correction unit 12 corrects and outputs the common information based on the correction information. On the other hand, when the waveform generation target voice quality of the voice waveform generation unit 16 is different from the target voice quality of the correction information, the common information correction unit 12 outputs the input common information as it is.
S34. The voice synthesis control command generation unit 14 converts the extracted common information into a voice synthesis control command for voice waveform generation.
S35. When the control command is input, the control command correction unit 15 checks whether the waveform generation target voice quality of the voice waveform generation unit 16 matches the target voice quality of the correction information.
S36. If they match, the control command correction unit 15 corrects the control command based on the separately input correction information. If they do not match, the control command correction unit 15 outputs the control command as it is.
S37. A speech waveform is generated by the speech waveform generation unit 16 based on the output information of the control command correction unit 15 and output as an output of the speech synthesizer 1.

  When the common information is composed of phonetic information and prosodic information for speech synthesis, by directly correcting the common information, the output synthesized speech can be greatly changed with small correction information. In this modification, since it is impossible to finely modify the sound that can be performed by modifying the command to the speech waveform generation unit 16, it is effective to combine these.

  In this case, for example, correction information is created by first creating correction information of common information, and correcting a command to the voice waveform generation unit 16 for a portion where a quality problem occurs in the synthesized voice. This is realized by means such as adding correction information.

  Also, a series of information for speech synthesis may be configured by combining one piece of common information and correction information for a plurality of types of voice qualities. In this case, in the voice waveform generation process, when a voice waveform is generated with a voice quality of the user's preference among a plurality of types of voice quality prepared in advance, the voice quality for which the correction information is prepared is more characteristic of the original voice. Even when there is no correction information, it is possible to output the voice of the target voice quality at a level possible with a conventional voice synthesizer. By creating correction information only for voice qualities that are expected to be popular, it is possible to efficiently meet the needs of the entire user while suppressing the overall data size.

  Moreover, all the embodiments described above are illustrative of the present invention and are not intended to limit the present invention, and the present invention can be implemented in other various modifications and changes. Therefore, the scope of the present invention is defined only by the claims and their equivalents.

DESCRIPTION OF SYMBOLS 1 Speech synthesizer 11 Pre-processing part 12 Common information correction part 13 Speech waveform synthesis part 14 Speech synthesis control command generation part 15 Control command correction part 16 Speech waveform generation part

Claims (7)

A speech synthesizer that synthesizes speech based on a series of speech synthesis information,
The speech synthesis information is composed of common information for synthesizing speech regardless of the target voice quality, and correction information that functions when voice synthesis is performed for a predetermined voice quality,
When synthesizing speech of a predetermined voice quality, correction means for changing the content of the common information based on the correction information;
Voice synthesis means for synthesizing a voice waveform based on the corrected information;
A speech synthesizer characterized by comprising: A speech synthesizer that synthesizes speech based on a series of speech synthesis information,
The speech synthesis information is composed of common information for synthesizing speech regardless of the target voice quality, and correction information that functions when voice synthesis is performed for a predetermined voice quality,
Control command generating means for generating a control command for generating a voice waveform based on the common information;
When synthesizing speech of a predetermined voice quality, control command correcting means for correcting the generated control command based on the correction information;
Voice waveform generation means for generating a voice waveform based on the modified control command;
A speech synthesizer characterized by comprising:   The speech synthesis apparatus according to claim 1, wherein the common information includes a phoneme symbol and a prosodic symbol.   4. The speech synthesizer according to claim 1, further comprising means for selecting a voice quality of the voice to be synthesized from a plurality of voice qualities. A speech synthesis method for synthesizing speech based on a series of speech synthesis information,
The speech synthesis information is composed of common information for synthesizing speech regardless of the target voice quality, and correction information that functions when voice synthesis is performed for a predetermined voice quality,
When synthesizing speech of a predetermined voice quality, a modification step for changing the content of the common information based on the modification information;
A speech synthesis method comprising a speech synthesis step of synthesizing a speech waveform based on the corrected information. A speech synthesis method for synthesizing speech based on a series of speech synthesis information,
The speech synthesis information is composed of common information for synthesizing speech regardless of the target voice quality, and correction information that functions when voice synthesis is performed for a predetermined voice quality,
A control command generating step for generating a control command for generating a voice waveform based on the common information;
When synthesizing speech of a predetermined voice quality, a control command correction step for correcting the generated control command based on the correction information;
A voice waveform generation step of generating a voice waveform based on the modified control command;
A speech synthesis method comprising:   A program that causes a computer to function as the speech synthesizer according to any one of claims 1 to 4.

Images