JP2006337476A - Voice synthesis method and system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice synthesis method for enhancing quality of synthesis voice by properly regulating rhythm, and to provide its system. <P>SOLUTION: The voice synthesis method selects prime pieces, determines whether rhythm modification is performed to the selected prime pieces, calculates a rhythm modification target value of the prime piece determining that the rhythm modification is performed on the basis of determination result, and performs the rhythm modification so that the rhythm of the prime pieces determining that the rhythm modification is performed is a calculated rhythm modification target value. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech synthesis method for synthesizing a desired speech.

  Conventionally, there is a speech synthesis technique for synthesizing a desired speech. This is realized by connecting speech segments corresponding to the desired utterance content and adjusting to have a desired prosody.

  Among speech synthesis technologies, one of typical technologies is based on a sound source-vocal tract model. This has a vocal tract parameter sequence as a speech segment. Using this vocal tract parameter, a synthesized speech can be obtained by filtering a pulse train simulating vocal cord vibration or noise simulating noise caused by expiration.

  On the other hand, recently, a speech synthesis technique called corpus-based speech synthesis has been widely used (for example, Non-Patent Document 1). In this method, voices with various variations are recorded in advance, and only the segments are connected during synthesis. In corpus-based speech synthesis, prosody adjustment is performed by appropriately selecting a segment having a desired prosody from a large number of segments.

In general, in corpus-based speech synthesis, compared to speech synthesis based on a sound source-vocal tract model, natural and high-quality synthesized speech can be obtained. This is said to be because it does not include processing such as modeling or signal processing that transforms speech (that is, causes deterioration of speech).
Seki, Miyagi, “Fragment selection method for high-quality speech synthesis using recorded speech of news programs,” IEICE Tech. 1-6, June (2003)

  However, the above corpus-based speech synthesis has a problem that the quality of synthesized speech is impaired when a segment having a desired prosody is not found. In particular, there is a problem that a prosodic gap is generated between a segment that deviates from a desired prosody and segments on both sides of the segment, which greatly impairs the naturalness of the synthesized speech.

  In order to solve the above problems, in the speech synthesis method according to the present invention, a step of selecting a segment, a step of determining whether or not to change the prosody for the selected segment, and a result of the determination Calculating the prosody change target value of the segment determined to be changed based on the prosody, and the prosody change of the segment determined to be changed to be the calculated prosody change target value The step of changing prosody is connected to the segment whose prosody has been changed and the segment which has been determined not to be changed as a result of the determination.

  According to the present invention, the prosody can be adjusted appropriately, and the quality of the synthesized speech can be improved.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows a hardware configuration in an embodiment of the present invention. Reference numeral 1 denotes a central processing unit that performs processing such as numerical calculation and control, and performs calculations according to the procedure of the present invention. Reference numeral 2 denotes an audio output unit that outputs audio. Reference numeral 3 denotes an input unit such as a touch panel, a keyboard, a mouse, or a button, which is used by the user to give an operation instruction to the apparatus. In the case of a device that operates autonomously regardless of a user instruction, the input unit may be omitted.

  Reference numeral 4 denotes an external storage unit such as a disk device or a non-volatile memory, which holds a language analysis dictionary 401 used for speech synthesis, a prosodic prediction parameter 402, a speech segment database 403, and the like. Further, the external storage unit 4 also holds information to be used permanently among various pieces of information held in the RAM 6. Further, the external storage unit 4 may be a portable form such as a CD-ROM or a memory card, which can enhance convenience.

  A read-only memory 5 stores program code 501 for realizing the present invention, fixed data (not shown), and the like. However, in the present invention, the use of the external storage unit 4 and the ROM 5 is optional. For example, the program code 501 may be installed in the external storage unit 4 instead of the ROM 5. Reference numeral 6 denotes a memory for holding temporary information such as a RAM, which holds temporary data, various flags, and the like. The central processing unit 1 to RAM 6 are connected by a bus 7.

  Next, the processing flow in the present embodiment will be described with reference to FIG. First, in step S1, the input speech synthesis target is analyzed. When the speech synthesis target is a natural language such as “Today is a good weather”, morphological analysis and syntax analysis, which are natural language processing techniques, are used in this step. At this time, the language analysis dictionary 401 is used as appropriate. On the other hand, when the speech synthesis target is described in an artificial language for speech synthesis, such as “Kyowa / Yo / I / Tenchides”, a dedicated analysis process is performed in this step.

  Next, in step S2, a phoneme sequence of the synthesized speech is determined based on the analysis result in step S1. Next, in step S3, a factor for selecting a segment is acquired based on the analysis result in step S1. As factors for selecting a segment, the number of mora and accent type of a phrase to which each phoneme belongs, the position in a sentence or phrase or word, the phoneme type, and the like are used.

  Next, in step S4, an appropriate segment is selected from the speech segment database 403 stored in the external storage unit 4 based on the result of step S3. As information for selecting a segment, not only the result of step S3 but also a segment can be selected so that the spectral shape and prosody gap between adjacent segments are reduced.

  Next, in step S5, the prosody value of the segment selected in step S4 is acquired. The prosodic value of the segment can be directly measured from the selected segment, or the value measured in advance can be stored in the external storage unit 4 and read out for use. In general, prosody often refers to fundamental frequency (F0), duration length, and power, but in this process, only a part of them may be acquired. This is because in the corpus-based speech synthesis, the prosody change is basically not performed, and it is not always necessary to acquire the ones that are not subject to the prosody change.

  Next, in step S6, the dissociation degree of the prosody value of the segment acquired in step S5 with respect to the prosody value of the adjacent segment is evaluated. If the dissociation degree is greater than the threshold value, the process proceeds to step S7 If not, the process proceeds to step S9.

  When the prosody is F0, a plurality of values correspond to one segment. For this reason, a plurality of methods are conceivable when evaluating the degree of dissociation. For example, representative values such as average value, intermediate value, maximum value, and minimum value of F0, F0 at the center of the segment, F0 at the end point of the segment, and the like can be used. Furthermore, the slope of F0 can be used.

  In step S7, the prosodic value after the prosody change is calculated. Most simply, a constant can be added and multiplied to the prosodic value of the segment acquired in step S5 so that the degree of dissociation used in step S6 is minimized. Alternatively, the prosody change amount can be limited to such an extent that the degree of dissociation used in step S6 falls within the threshold range. Furthermore, the prosodic value after the prosody change can be obtained by interpolating the prosodic value of the segment determined not to change the prosodic value in step S6.

  Next, in step S8, the prosody of the segment is changed based on the prosody calculated in step S7. As a prosody changing method, various methods (for example, PSOLA method) used in conventional speech synthesis can be used.

  In step S9, based on the result of step S6, the segment selected in step S4 or the segment whose prosody was changed in step S8 is connected and output as synthesized speech.

  With the above configuration, even if a segment having a desired prosody is not found, there is a prosodic gap between the segment that is out of the desired prosody and the segments on both sides thereof. Alleviated. As a result, the quality of the synthesized speech is prevented from being greatly impaired.

  As a second embodiment, there is a method in which prosody prediction is performed based on the language analysis result and the predicted prosody is used.

  The processing flow in the present embodiment will be described with reference to FIG. First, step S1 and step S2 are the same processes as in the first embodiment. Next, in step S101, a factor for predicting the prosody is acquired based on the analysis result in step S1. As factors for predicting the prosody, the number of mora and accent type of the phrase to which each phoneme belongs, the position in the sentence or phrase or word, the phoneme type, information on the adjacent phrase or word, etc. are used.

  Next, in step S102, the prosody is predicted based on the prosody prediction factor acquired in step S101 and the prosody prediction parameter 402 stored in the external storage unit 4. In general, prosody often refers to fundamental frequency (F0), duration length, and power, but in this process, only a part of them may be predicted. This is because in corpus-based speech synthesis, the prosody of the segment can be used, so it is not always necessary to predict all information.

  Next, in the same manner as in the first embodiment, segment selection factor acquisition (step S3), segment selection (step S4), and segment prosody acquisition (step S5) are performed. In this embodiment, since prosody prediction is performed, the prosody predicted in step S102 can be used as information for selecting a segment.

  Next, in step S103, the degree of dissociation between the prosodic value predicted in step S102 and the prosody value of the segment acquired in step S5 is evaluated. If the degree of dissociation is greater than a threshold value, processing is performed. The process proceeds to S104. Otherwise, the process proceeds to step S9.

  When the prosody is F0, a plurality of values correspond to one segment. For this reason, a plurality of methods are conceivable when evaluating the degree of dissociation. For example, representative values such as the average value, intermediate value, maximum value, and minimum value of F0, F0 at the center of the segment, and the mean square error between the predicted prosodic value and the prosodic value of the segment can be used.

  Next, in step S104, the prosodic value after the prosody change is calculated. Most simply, the prosodic value after the prosody change can be made equal to the prosodic value predicted in step S102. Alternatively, the prosody change amount can be limited to such an extent that the degree of dissociation used in step S103 falls within the threshold range. Steps S8 and S9 are the same as those in the first embodiment.

  As a third embodiment, there is an example in which the segment selection is redone for the segment determined to be changed in the above-described embodiment.

  The processing flow in the present embodiment will be described with reference to FIG. First, steps S1 to S5 are processed in the same manner as in the second embodiment. Next, in step S103, the same determination as in the second embodiment is performed. If the degree of dissociation is larger than the threshold value, the process proceeds to step S201. Otherwise, the process proceeds to step S9.

  In step S201, a factor for reselecting the segment is acquired. As a factor for reselecting the segment, not only the factor used in step S3 but also information on the segment determined not to change the prosody in step S103 can be used. For example, the continuity of the prosody can be improved by using the segment prosody value of the segment determined not to be changed in step S103. Alternatively, the continuity of the spectrum with the segment acquired in step S5 can be considered.

  Next, in step S202, as in step S4, a segment is selected based on the result of step S201.

  Further, as in the first embodiment, when the dissociation of prosodic values between adjacent segments is large, prosody change is performed (step S6, step S7, step S8). Finally, the synthesized sound is output in the same manner as in the previous embodiment (step S9).

  By setting it as such a structure, it becomes possible to select a more suitable segment.

  An object of the present invention is to supply a storage medium recording a program code of software that realizes the functions of the above-described embodiments to a system or apparatus, and the computer (or CPU or MPU) of the system or apparatus stores the storage medium. Needless to say, this can also be achieved by reading and executing the program code stored in.

  In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the storage medium storing the program code constitutes the present invention.

  As a storage medium for supplying the program code, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a CD-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

  Further, by executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also an OS (operating system) operating on the computer based on the instruction of the program code. It goes without saying that a case where the function of the above-described embodiment is realized by performing part or all of the actual processing and the processing is included.

  Further, after the program code read from the storage medium is written into a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion is performed based on the instruction of the program code. It goes without saying that the CPU or the like provided in the board or the function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing.

It is the block diagram which showed the hardware constitutions in the Example of this invention. 3 is a flowchart illustrating a processing flow in the first embodiment. 10 is a flowchart illustrating a processing flow in the second embodiment. 10 is a flowchart illustrating a processing flow in Embodiment 3.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Central processing part which performs processing, such as numerical calculation, control 2 Audio | voice output part which presents audio | voice to a user 3 Input parts, such as a touch panel, a keyboard, a mouse | mouth, and a button 4 External storage parts, such as a disk apparatus and a non-volatile memory 5 Reading Dedicated memory 6 Memory for holding temporary information such as RAM 7 Bus 401 Language analysis dictionary used for speech synthesis 402 Prosody prediction parameter used for speech synthesis 403 Speech unit database 501 used for speech synthesis Program code to realize

Claims (10)

Selecting a fragment;
Determining whether to change the prosody for the selected segment;
Calculating a prosody change target value of a segment determined to perform prosody change based on the determination result;
Changing the prosody so that the prosody of the segment determined to perform the prosody change becomes the calculated prosody change target value;
A speech synthesizing method comprising connecting the segment whose prosody has been changed and the segment which has been determined not to be changed as a result of the determination.   The speech synthesis method according to claim 1, wherein whether or not to change the prosody is determined based on a prosody dissociation degree between adjacent segments.   The speech synthesis method according to claim 1, wherein the prosodic change target value is calculated based on a prosodic value of a segment determined not to change prosody as a result of the determination. Further comprising the step of predicting the prosody,
The speech synthesis method according to claim 1, wherein whether or not to change the prosody is determined based on a degree of dissociation from the predicted prosody. Further comprising predicting prosody,
The speech synthesis method according to claim 1, wherein the prosodic change target value is calculated based on the predicted prosodic value.   The speech synthesis method according to claim 1, further comprising a step of reselecting a segment with respect to a segment determined to change prosody as a result of the determination.   The speech synthesis method according to claim 6, wherein the reselection of the segment is performed based on information of a segment determined not to change the prosody as a result of the determination.   The method further comprises a step of re-determining whether or not a prosody change is performed on the reselected segment, and the prosody change is performed on the re-determined segment when the prosody change is performed. Item 7. The speech synthesis method according to Item 6.   A control program for causing a computer to execute the speech synthesis method according to claim 1. Means for selecting the pieces;
Means for determining whether or not to change the prosody for the selected segment;
Means for calculating a prosody change target value of a segment determined to perform prosody change based on the determination result;
Means for changing the prosody such that the prosody of the segment determined to perform the prosody change becomes the calculated prosody change target value;
A speech synthesizer characterized by connecting the segment whose prosody has been changed and the segment which has been determined not to be changed as a result of the determination.

Images