JP2008268477A - Rhythm adjustable speech synthesizer - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a user friendly speech synthesizer equipped with a graphic user interface (GUI) for adjusting rhythm. <P>SOLUTION: The speech synthesizer is equipped with: an intermediate language creation section 24 for creating an intermediate language to which a rhythm parameter is attached, from an input text data; an editing image creating section 27 for creating an editing image which can be edited, by graphing a value of the rhythm parameter which is attached to the intermediate language; a rhythm parameter editing section 23 which rewrites the value of the rhythm parameter to an indicated value by displacing a display symbol displayed on a graph of the editing image via a setting receiving section 22; a speech synthesis section 25 for synthesizing a waveform based on the rewritten rhythm parameter; and a speech output section 26 for outputting the synthesized waveform via an output device. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a text-to-speech synthesis technology for synthesizing speech from text, and more particularly to a technology for adjusting prosody such as a fundamental frequency and a duration length.

  In recent years, in a system for synthesizing speech from text, in order to obtain higher-quality speech, there is a method of using a speech corpus in which speech segments (speech waveform fragments) are cut out from natural speech and the features are accumulated. Are known. The features of speech include prosodic parameters such as a fundamental frequency indicating the pitch and a phoneme duration indicating the length of each phoneme. A natural speech is synthesized by extracting speech segments from the speech corpus in which these are stored using an arbitrary criterion.

  However, such a conventional method selects and automatically corrects a prosodic parameter that is assumed to be optimal from preset prosodic parameters based on a related phrase selected by the user. Therefore, even if a related phrase is specified, it may not be corrected to the expected pronunciation. The same applies when there is no relevant word / phrase in the co-occurrence data.

  Patent Document 1 proposes a speech synthesizer that includes an editing unit that allows a user to adjust a prosodic parameter, for example, an intermediate language that includes information on duration and fundamental frequency. This is because a phrase related to a phrase to be corrected is stored in advance as co-occurrence data together with its reading and prosodic information, and a list display of the related co-occurrence data is displayed using a UI (User Interface). It is possible. When the user selects an optimum related phrase from among them, the reading and inflection of the phrase to be corrected is corrected to be more natural.

JP 2006-30326 A

  However, the conventional method as described above selects and automatically corrects a prosodic parameter that is assumed to be optimal from preset prosodic parameters based on a related phrase selected by the user. Therefore, even if a related phrase is specified, it may not be corrected to the expected pronunciation. The same applies when there is no relevant word / phrase in the co-occurrence data.

  In the present invention, in order to solve the above-described problems of the prior art, the prosody parameters can be visually recognized and operated on the screen, and the user who has no specialized knowledge can easily perform the prosody parameter editing operation. An object is to provide a friendly GUI (Graphical User Interface).

  In order to solve the above problems, the present invention acquires a prosodic parameter for the input character string from a speech corpus, and associates the prosodic parameter with the input character string for each prosodic control unit, An intermediate language generating means for generating an intermediate language, and a first prosodic parameter corresponding to the horizontal axis and a second prosodic parameter corresponding to the vertical axis are defined from the prosodic parameters included in the generated intermediate language. Edit the prosodic parameters by displaying a pre-stored display symbol on the graph and displaying it on the screen of the connected display device at the coordinates for each prosodic control unit of the intermediate language. There is provided a speech synthesizer comprising: a prosodic parameter editing screen generating means for generating a screen.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
First, an outline of an embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing a hardware system configuration of a speech synthesizer according to claim 1 of the present invention.

  As shown in FIG. 1, the speech synthesizer 10 is a general computer on which a program operates, for example, a personal computer or a workstation. That is, the speech synthesizer 10 includes a CPU (Central Processing Unit) 1 that is a main part of a computer and controls each device in a centralized manner, and a main storage device 2 that stores various data in a rewritable manner.

  Furthermore, the speech synthesizer 10 includes an external storage device 3 that stores various programs, data generated by the programs, an input device 4 such as a keyboard and a mouse for performing various operation instructions, and a display device that displays image data and the like. 5. An output device 6 for outputting audio data and the like as audio is provided. Each of these devices is connected to the CPU 1 through a signal line 7 such as a bus. Of course, in addition, a communication device for communicating with an external device may be provided. The external storage device 3 includes, for example, an HDD (Hard Disk Drive).

  The CPU 1 executes various processes by, for example, loading a program stored on the external storage device 3 onto the main storage device 2 and executing it. The external storage device 3 is not limited to the HDD, and may further include a drive such as a CD-ROM or a DVD-ROM as a mechanism for reading computer software that is a distributed program. Of course, the program may be downloaded from the network to the external storage device 3 via a communication device, and then loaded onto the main storage device 2 and executed by the CPU 1.

  The input device 4 includes a text input device for inputting text, and a pointing device for operating graphics indicating a target operation on the GUI. The text input device may be, for example, a keyboard, a voice recognition device, or a character string reading device as long as it can input a character string. The pointing device may be, for example, a mouse or a touch panel that directly contacts the screen.

  The display of the display device 5 is selected from CRT (Cathode Ray Tube), LCD (Liquid Crystal Display), and the like.

  The output device 6 may be an external output device such as an external speaker as long as it converts audio data sent from the CPU into audio and outputs the audio.

  FIG. 2 is a block diagram showing a functional configuration of the speech synthesizer 10 composed of the above hardware. In the present embodiment, the speech synthesizer 10 is assumed to have a basic GUI editing function.

  As shown in the figure, a control unit 20 and a storage unit 30 are constructed on the speech synthesizer 10. The control unit 20 includes a setting reception unit 22, a prosodic parameter editing unit 23, an intermediate language generation unit 24, a speech synthesis unit 25, a speech output unit 26, and an editing screen generation unit 27. The storage unit 30 includes a dictionary data storage area 32, a speech corpus storage area 33, and a work data storage area 34.

  These functions are performed, for example, by the CPU 1 loading a predetermined program stored in advance in the auxiliary storage device 3 into the main storage device 2 and executing it, or by controlling the hardware, or a combination thereof. It is realized by. The storage unit 30 is realized by using the external storage device 3 when holding data continuously, and using the main storage device 2 when holding data temporarily.

  The setting accepting unit 22 accepts an operation by a user on the GUI, for example, input of text data, movement of a cursor / pointer, and the like via the input device 4.

  The prosody parameter editing unit 23 receives an operation related to prosody parameter editing via the setting reception unit 22. Then, the change of the position of the display symbol associated with the prosodic parameter is detected, the value of the prosodic parameter is calculated from the displacement amount and the displacement direction, and the new prosodic parameter is given to the intermediate language.

  The intermediate language generation unit 24 receives the input text data via the setting reception unit 22. Then, the received text data is divided into words based on the dictionary data 700 and the like, and morphological analysis data including reading / accent information and accent phrase information is generated, and the input text data has prosody. Similar data is searched and extracted from the speech corpus 400 stored in the external storage device. Further, based on this, a process for generating reference language parameters and generating intermediate language data is performed.

The speech synthesizer 25 synthesizes the output speech waveform based on the intermediate language data generated by the intermediate language generator 24 to generate synthesized waveform data.
The audio output unit 26 outputs the generated synthesized waveform data as actual audio data via the output device 6.

  The edit screen generation unit 27 graphs the prosodic parameter information included in the generated intermediate language, generates a screen for the user to perform speech synthesis work, for example, a text input screen or a prosody parameter edit screen, and displays the display device 5 is displayed.

  The dictionary data storage area 32 stores in advance a dictionary 700 that stores word readings, accent information, and the like.

  As shown in FIG. 5, the speech corpus storage area 33 stores information in which words, accent phrases, and character string groups in phrase units are associated with prosodic parameters such as fundamental frequency and duration, speech data, and the like in advance. Stores speech corpora stored as a database. Specifically, for example, as shown in FIG. 5, character string notation data 410, speech waveform data 420 that is the utterance of the character string notation data 410, basic frequency data 430 of the speech waveform data 420, and speech waveform data 420 Data set (4100, 4200... N) composed of duration time data 440, morpheme segmentation data 450 which is a morpheme segmentation result of character string representation data 410, and phoneme segmentation data 460 which is a phoneme segmentation result of character string representation data 410 Are provided. The contents of the constituent elements are not limited to the above, and may include power data, cepstrum data, and the like.

  The work data storage area 34 is an area for temporarily storing input text data, intermediate data generated by the prosodic parameter editing unit 23 and the intermediate language generation unit 24, and the like. Specifically, as shown in FIG. 3, text data 610, morpheme analysis data 620, phoneme analysis data 630, reference prosody parameter 640, updated prosody parameter 650, search result data 660, intermediate language data 670, intermediate language update data This is an area for storing 680 various data.

  Next, the operation of the speech synthesizer configured with the above-described functions will be described with reference to a flowchart.

  FIG. 4 is a flowchart showing the flow of processing from receiving text data input to performing waveform synthesis. Here, the description will be made taking text data “raining” as an example.

  First, the setting accepting unit 22 accepts text data to be synthesized (S1).

  Specifically, the setting reception unit 22 displays a text input screen shown in FIG.

  Here, the configuration of the text input screen (FIG. 8) will be described. The text input screen includes a prosody edit button 601 and an input text setting field 602. The prosody editing button 601 is a button for starting prosody parameter editing. The input text setting field 602 is a field for setting text data to be subjected to speech synthesis.

  After displaying the text input screen, the setting reception unit 22 receives a user operation on the text input screen via the input device 4. When text data is input to the input text setting field 602, the setting receiving unit 22 stores the text data in the text data 610 on the work data storage area 34. That is, the text data “rains” is stored in the text data 610 (S1).

  Of course, the method of accepting text data is not limited to the above. For example, the setting reception unit 22 displays a GUI screen that receives input of text data of a sentence composed of a plurality of sentences in the previous stage of the above text input screen. The text data of the input sentence is displayed on the screen, and the user is allowed to select one sentence for prosody adjustment, and then the above text input screen is displayed to display the selected sentence. can do.

  Next, when the input text data is set, the intermediate language generation unit 24 generates an intermediate language from the input text (S2).

  Here, the intermediate language generation process (S2) will be described in detail with reference to FIG.

  The intermediate language generation unit 24 reads the text data “rains” stored in the text data 610 on the work data storage area 34 (S101).

  Next, the intermediate language generation unit 24 executes morpheme analysis processing (S102). Specifically, the morpheme analyzer 24 divides the read text data into the smallest units (morphemes) whose meaning is known. Then, for each divided morpheme, morpheme analysis data composed of notation, reading, accent information, etc. is generated, and the data is stored in the morpheme analysis data 620 in the work data storage area 34. As the reading and accent information for each word, values registered in the dictionary 700 in advance are used. As a method of dividing the text data into words (morphemes), Shimizu et al. “Text-to-speech morpheme analysis processing focusing on the connection relationship between adjacent words, Journal of the Acoustical Society of Japan, Vol. 51, No. 1, pp. 3- 13, 1995 "can be used. Of course, this method is an example, and other processing methods may be used.

  As described above, the intermediate language generation unit 24 generates morphological analysis data as shown in FIG. 13B from the text data “rains down” in FIG. That is, the data for each word is divided into “rain”, “ga”, “fall”, “ru”, “.”, And reading / accent information “A'me”, “ka”, “fu”, “le”, “.” (“′” "Is an accent, and" ° "is a nasal cloud sound). Also, information “/” indicating an accent phrase delimiter is added. “A'mecha / full” corresponds to a phonetic symbol string. Of course, the structure of the morpheme is not limited to the above.

  Next, the intermediate language generation unit 24 executes phoneme division processing (S103). First, the morpheme analysis data 620 is read, and based on the reading information included in the data, the phoneme analysis data is divided into the minimum sound units (phonemes) used for meaning distinction, and the obtained phoneme analysis data is stored in a work data storage area. 34 is stored in phoneme analysis data 630.

  The phoneme segmentation method uses, for example, the method by Miyazaki et al. “Language Processing Method for Japanese Sentence Speech Output, Journal of Information Processing Society of Japan, Vol. 27, No. 11, pp. 1053-1061, 1986”. Can do. Of course, this calculation method is an example, and other phoneme division methods may be used.

  By the above phoneme analysis, the intermediate language generation unit 24 divides the text data “rains down” into phonemes, and displays “A / ME / NG / A / H / U /” shown in FIG. Phoneme data such as “R / U /.” Is generated. Here, “A”, “M”, “E”, and the like are symbols indicating phonemes, but these are only examples, and other phoneme symbol expressions may be used.

  Next, the intermediate language generation unit 24 performs processing for searching the speech corpus 400 for a data set having similar information such as accent type and part of speech with respect to the morphological analysis data (S104 to S106).

  First, the intermediate language generation unit 24 reads the data set 4100 from the speech corpus 400 (S104). Further, the morpheme division data 450 is read from the read data set 4100.

  Then, the intermediate language generation unit 24 reads the morpheme analysis data 620 (S105), compares it with data such as reading / accent information, accent type, part of speech, etc. included in the morpheme division data 450, and compares the similarity with a predetermined criterion. Is calculated (S106).

  Thereafter, similarly, the intermediate language generation unit 24 performs the above similarity calculation for all the data sets (4200 to n) (S106). As a result, one data set that is most similar to the morphological analysis data (hereinafter referred to as a selection data set) is selected from among the data sets that satisfy a preset threshold value (reference similarity).

  Next, the intermediate language generation unit 24 calculates prosodic parameters for the morphological analysis data 620 (S107).

  Specifically, the intermediate language generation unit 24 compares the morpheme analysis data 620 and the morpheme division data 450 of the selected data set, and separates the morpheme into a match part and a mismatch part. The prosody parameters (basic frequency data 430 and duration time data 440) of the selected data set are assigned to the morphemes of the matching part. The fundamental frequency data of the morpheme of the mismatched part is calculated by searching from a word fundamental frequency pattern table storing one fundamental frequency data for the number of mora and accent type of the morpheme. Also, the duration time is “Sonic time length control for speech synthesis by rules, IEICE Transactions, Vol. J67-A, No. 7, pp. 629-636, 1984” by Osaka et al. It can be calculated using. Thereafter, the intermediate language generation unit 24 performs a process of transforming and integrating the mismatched parts in order to smoothly connect the prosodic parameters of the matched part and the mismatched part.

  The intermediate language generation unit 24 stores the prosodic parameters thus obtained in the reference prosodic parameters 640 in the work data storage area 34. The method for calculating the prosodic parameters is not limited to the above. In order to obtain the duration time for each phoneme, for example, a table in which a duration time for each phoneme is recorded in advance as a database, or a duration considering environmental factors up to one phoneme before and after the target phoneme Can be used to refer to the table in which is recorded. In addition, in order to obtain the fundamental frequency for each phoneme, a method of modeling with an exponential curve called a second critical braking model or a method of modeling with a rectangle can be used.

  Next, the intermediate language generation unit 24 generates intermediate language data based on the morpheme analysis data 620, the phoneme analysis data 630, and the reference prosody parameter 640, and stores the intermediate language data in the intermediate language data 670 on the work data storage area 34. Store (S108).

  Specifically, the intermediate language generation unit 24 generates intermediate language data as shown in FIG. That is, the phonetic symbol string included in the reference morphological analysis data is divided to generate a data string composed of phoneme notations of “a”, “me”, “cap”, “/”, “fu”, “le”, and “.”. And the data of a fundamental frequency and duration length are provided for every phoneme of each character. For example, “me” includes a phoneme “M” having a fundamental frequency “283” and a duration “51”, and a phoneme “E” having a fundamental frequency “252” and a duration “89”.

  As described above, the intermediate language generation unit 24 ends the intermediate language generation process (S2).

  Next, the prosody parameter editing process (S3) will be described with reference to FIG.

  First, the setting receiving unit 22 receives a click on the prosody editing button 601 on the text input screen (FIG. 8) (S310). Then, the edit screen generation unit 27 starts an edit screen generation process.

  The edit screen generation unit 27 generates a prosody parameter edit screen 800 as shown in FIG. 9 having user-variable display symbols (S312). First, the intermediate language data 670 is read from the work data storage area 34 (S311). Then, phoneme notation 910 and phoneme notation 920 are extracted (see FIG. 13D). First, the phoneme notation 910 is expanded in the horizontal axis direction as a phoneme notation character string 830 (FIG. 9), and the corresponding phoneme notation 920 is further expanded as a phoneme notation character string 840.

  Next, the edit screen generation unit 27 extracts a duration length parameter 930 and a fundamental frequency parameter 940 from the intermediate language data 670.

  Further, the expanded phoneme notation character string 830 and the phoneme notation character string 840 are associated with the duration parameter 930 on the horizontal axis and the fundamental frequency parameter 940 on the vertical axis, and the prosodic parameters are graphed. Execute.

Next, the edit screen generation unit 27 sets the duration length display symbol 850 extending in the vertical axis direction at the coordinate of the horizontal axis (duration duration parameter 930), which is the starting point for each character of the phoneme-notation character string 840. The phoneme-notation character string 840, which is a unit for adjusting the continuation length, is arranged so as to delimit characters. (For convenience, a symbol is attached only to one symbol. The same applies hereinafter.)
The editing screen generation unit 27 sets the fundamental frequency display symbol 860 on the duration display symbol 850 arranged on the horizontal axis coordinate serving as the starting point for each character of the phoneme notation character string 830, and the value of the fundamental frequency parameter. Arrange according to. Further, a line connecting adjacent fundamental frequency display symbols 860 is generated as a prosodic connection symbol 870.

  Here, the duration length display symbol 850 is in advance in the horizontal axis direction indicated by the arrow line X in FIG. 9, and the fundamental frequency display symbol 860 is in advance in the vertical axis direction indicated by the arrow line Y in FIG. Within the range of the set reference value, a sliding operation via the input device 4 is possible. The duration length display symbol 850 can change the width with the adjacent 850 on the left side, and can increase or decrease the duration length of the corresponding written character. The fundamental frequency symbol 860 can be slid on the duration length display symbol 850, and the fundamental frequency of the corresponding written character can be increased or decreased. The arrow lines X and Y in the figure do not indicate the operation range, but merely indicate the operation direction.

  In the edit screen generation unit 27, the horizontal axis has a duration of milliseconds (1 pixel = 1 ms) and the vertical axis has a frequency of Hz (minimum frequency of target data × 0.8 to maximum frequency ÷ 0.75). The prosody editing screen 800 is generated. In addition, although such a unit was used here, of course, you may produce | generate a graph using another unit.

  In addition, when the prosody parameter editing screen 800 does not fit within the screen of the display device, a scroll bar is configured so that the screen can be slid left and right. Note that the screen switching means is not limited to the scroll bar, and may be provided with functions for switching pages and compressing and displaying the whole.

  The editing screen generation unit 27 displays the prosody parameter editing screen 800 generated as described above on the display device 5 using the GUI.

  The setting receiving unit 22 receives a display symbol moving operation (S321).

  First, the setting reception unit 22 detects that the display symbol has been moved through the input device 2.

  When the click operation with the mouse, which is the pointing device of the input device 4, is detected on the duration length display symbol 850, the setting reception unit 22 starts accepting the slide operation by dragging in the arrow X direction of 850. To do. Next, when a sliding operation in the direction indicated by the arrow X is detected, the prosodic parameter editing unit 23 acquires the displacement amount of the duration display symbol 850 after the slide and information about the sliding direction. Further, a rewrite value determined by the sliding direction and the displacement amount is calculated based on the unit of the horizontal axis coordinate.

  Similarly, regarding the change of the fundamental frequency, when a click operation with the mouse is detected on the fundamental frequency display symbol 860, the prosody parameter editing unit 23 starts accepting a slide operation by dragging in the direction indicated by the arrow Y in 860. Next, when a slide operation in the direction indicated by the arrow Y is detected, the displacement amount of the fundamental frequency display symbol 860 after the slide and information about the slide direction are acquired. Further, the rewriting determined by the sliding direction and the displacement amount is calculated based on the unit of the vertical axis coordinate.

  Here, the operation via the pointing device is performed using a mouse here, but of course, a touch action on the touch panel or the like may be used.

  With reference to FIG. 10, the calculation process of the rewrite value of the prosodic parameter will be specifically described. The arrow line A → A ′ indicates the displacement of the duration length display symbol 850. First, the setting reception unit 22 receives a slide operation in the arrow direction A → A ′ via the pointing device. Then, the prosody parameter editing unit 23, the duration display symbol 850 to be operated, and all the phoneme notation, phoneme notation, and display symbols (the duration length display symbol 850, the basic frequency, and the subsequent horizontal coordinate). The display symbol 860 and the prosodic connection symbol 870) are displayed with the same displacement amount and displacement in the slide direction of the arrow line A → A ′.

  Here, in the slide operation from A → A ′, the slide direction of the X-axis coordinate is −, so the displacement is represented by − (A′−A) (the unit is the pixel value of the X-axis coordinate). . Therefore, in this case, since one pixel is set to 1 millisecond, the prosody parameter editing unit 23 sets the millisecond equivalent to the pixel displacement amount of the X-axis coordinate to the duration time (A Subtract from (A0) and calculate the rewrite value of the duration parameter.

  Further, the arrow line B → B ′ indicates the displacement of the fundamental frequency display symbol 860. First, the setting reception unit 22 receives a slide operation in the arrow direction B → B ′ via the pointing device. Then, the prosodic parameter editing unit 23 displays the operation target fundamental frequency display symbol 860 by displacing it to the B ′ coordinate. According to the displacement of 860, the prosodic connection symbol 870 is also displaced simultaneously.

  Here, in the slide operation of B → B ′, since the slide direction of the Y-axis coordinates is +, the displacement amount is represented by (B′−B) (the unit is the pixel value of the Y-axis coordinates). The prosodic parameter editing unit 23 uses a coefficient (Hz / Hz) representing a frequency corresponding to one pixel determined by the Y-axis range of the graph (the minimum frequency of the target data × 0.8 to the maximum frequency ÷ 0.75). pixel) to calculate the fundamental frequency value. This is added to the fundamental frequency B of the corresponding phoneme notation, and the rewritten value of the fundamental frequency parameter is calculated.

  The prosodic parameter editing unit 23 determines that the rewriting values of the parameters of the duration time, the fundamental frequency, and the phrase duration time length calculated from the above processing are within a predetermined reference value range. Confirm.

  The prosodic parameter editing unit 23 displays an error screen on the screen when the rewritten value is outside the predetermined range (S145). For example, an error message such as “Set the frequency within ~ Hz” or “Set the duration to ~ ms or more” is displayed on the error screen.

  At the same time as displaying the error message about the duration, the prosodic parameter editing unit 23 coordinates the upper limit value when the rewritten value exceeds the upper limit value, and coordinates the lower limit value when the rewrite value exceeds the lower limit value. In addition, processing for displacing the duration length display symbol 850 is executed. For the fundamental frequency, a process of displacing the fundamental frequency display symbol 860 to the coordinates immediately before the slide operation exceeding the reference value, which is a cause of displaying an error message, is executed.

  When the rewritten value is within the reference value, the prosodic parameter editing unit 23 stores the rewritten value as the updated prosodic parameter 650 in the work data storage area 34. Further, a rewriting process is performed in which the prosodic parameters of the intermediate language data 670 are rewritten to the values of the updated prosodic parameters 650 (S324). Thus, the updated intermediate language data is stored as intermediate language update data 680 on the work data storage area 34 (S325).

  From the intermediate language update data 680 generated by the prosodic parameter editing process executed as described above, the speech synthesizer 25 executes waveform synthesis (S4). Further, the synthesized waveform is output by the audio output unit 26 via the output device 6.

  In this embodiment, the fundamental frequency can be changed only for each phoneme notation, that is, for each vowel start frequency, and the consonant start frequency is automatically set according to the value of the vowel start frequency. You may comprise the edit screen which can set a fundamental frequency (refer FIG. 11).

  Further, an editing screen may be configured in which a fundamental frequency editing point 880 that can edit the fundamental frequency is provided in the phoneme notation (see FIG. 12). Further, a function may be provided in which the user can freely specify the coordinates in the phoneme notation and arrange the fundamental frequency editing point 880 at an arbitrary position.

  The first embodiment has been described above. According to the first embodiment, a prosodic parameter editing screen is provided on which prosody parameters can be visually recognized and edited on a graph. As a result, even a user with poor expertise can adjust the prosody visually and intuitively with a simple operation. Conversely, even users who have specialized knowledge can specify the values of prosodic parameters specifically, and are not limited to predetermined prosodic patterns. Thus, the present invention can improve usability.

  The present invention has been described in connection with exemplary embodiments. Obviously, many alternatives, modifications, and variations will be apparent to practitioners skilled in this art. Accordingly, the above-described embodiments of the present invention are intended to illustrate and not limit the spirit and scope of the present invention.

The block diagram which shows the structure of the speech synthesizer of 1st embodiment. The block diagram which shows the function structure of the speech synthesizer of 1st embodiment. Schematic which shows the structure of a work data storage area. The flowchart which shows the whole process which the speech synthesizer of 1st embodiment performs. Schematic which shows the structure of an audio corpus. Flow chart explaining intermediate language generation processing The flowchart explaining prosody parameter edit processing. Schematic which shows the example of a display of a text input screen. Schematic which shows the example of a display of a prosodic parameter edit screen. Schematic which showed the example of a display of the prosodic parameter edit screen which receives an edit process. Schematic which showed the example of a display of the prosodic parameter edit screen which enabled the fundamental frequency to be designated for every phoneme. Schematic which showed the example of a display of the prosodic parameter edit screen which further provided the point which can designate a fundamental frequency in phoneme. Explanatory drawing which shows an example of morpheme analysis data, phoneme analysis data, and intermediate language data.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Speech synthesizer, 1 ... CPU, 2 ... Main storage device, 3 ... External storage device, 4 ... Input device, 5 ... Display device, 6 ... Output device, 7 ... Bus 20 ... Control part, 22 ... Setting reception part , 23 ... Prosodic parameter editing section, 24 ... Intermediate language generation section, 25 ... Speech synthesis section, 26 ... Speech output section, 27 ... Editing screen generation section 30 ... Storage section, 32 ... Dictionary data storage area, 33 ... Speech corpus storage Area 34: work data storage area 700 dictionary 610 text data 620 morphological analysis data 630 phoneme analysis data 640 reference prosody parameter 650 update prosody parameter 660 search result data 670 intermediate Language data, 680 ... Intermediate language update data 400 ... Speech corpus, 4100, 4200 ... Data set, 410 ... Character string notation data, 420 ... Speech waveform data 430 ... fundamental frequency data, 440 ... duration length data, 450 ... morpheme division data, 460 ... phoneme division data 601 ... prosody edit button, 602 ... input text setting field 800 ... prosody parameter edit screen, 830 ... phoneme description character string , 840 ... Phoneme notation character string, 850 ... Duration length display symbol, 860 ... Fundamental frequency display symbol, 870 ... Prosodic connection symbol, 880 ... Fundamental frequency edit point 910 ... Phoneme notation, 920 ... Phoneme notation, 930 ... Duration length Parameter, 940 ... fundamental frequency parameter

Claims (9)

A speech synthesizer that synthesizes speech corresponding to an input character string,
Prosody parameters including at least information specifying accent, duration length, and fundamental frequency as parameters, and voice data, corresponding to at least one of phoneme display characters and phoneme display characters as prosodic control units Storage means for storing the voice corpus accumulated for each voice control unit;
An intermediate language generating means for dividing the input character string into prosodic control units and generating an intermediate language in which prosodic parameters are associated with each divided prosodic control unit;
Among the parameters included in the prosodic parameters, a graph including coordinates in which the values of the first parameter and the second parameter are respectively arranged on the horizontal axis and the vertical axis is formed, and the first parameter and each of the divided prosodic control units Prosody parameter editing screen generating means for generating a prosodic parameter editing screen in which a predetermined display symbol is displayed on the graph at the coordinate position specified by the second parameter, and displaying on the display means;
A speech synthesizer comprising: The speech synthesizer according to claim 1,
In the prosodic parameter editing screen displayed on the display means, a displacement of the coordinate position of the display symbol is received via the input means, and the prosodic parameter of the intermediate language corresponding to the prosodic control unit in which the display symbol is displaced Prosody parameter rewriting means for changing the value of the prosody parameter value to a prosodic parameter value specified from the coordinate position of the display symbol after displacement
A speech synthesizer further comprising: The speech synthesizer according to claim 1,
The first parameter is a duration length and the second parameter is a fundamental frequency;
A speech synthesizer characterized by the above. The speech synthesizer according to claim 1 or 3,
The prosodic parameter editing screen generation means includes:
Generating a prosodic parameter editing screen that further expands and displays a phoneme notation character string, a phoneme notation character string, or both as a phoneme control unit in the horizontal axis direction;
A speech synthesizer characterized by the above. The speech synthesizer according to claim 3 or 4,
The prosodic parameter editing screen generation means includes:
A first display symbol representing a duration parameter corresponding to the horizontal axis, a second display symbol representing a fundamental frequency parameter corresponding to the vertical axis, and a connection connecting the adjacent second display symbols to each other; A prosody parameter editing screen that displays the corresponding to the prosody control unit,
The first display symbol is arranged at a coordinate corresponding to a duration length parameter value, and the second display symbol is arranged at a coordinate corresponding to a fundamental frequency parameter value on the first display symbol. ,
A speech synthesizer characterized by the above. A program that causes a computer to function as a speech synthesizer that synthesizes speech corresponding to an input character string,
The computer,
Prosody parameters including at least information specifying accent, duration length, and fundamental frequency as parameters, and voice data, corresponding to at least one of phoneme display characters and phoneme display characters as prosodic control units Storage means for storing the voice corpus accumulated for each voice control unit;
An intermediate language generating means for dividing the input character string into prosodic control units and generating an intermediate language in which prosodic parameters are associated with each divided prosodic control unit;
Among the parameters included in the prosodic parameters, a graph including coordinates in which the values of the first parameter and the second parameter are respectively arranged on the horizontal axis and the vertical axis is formed, and the first parameter and each of the divided prosodic control units Prosody parameter editing screen generation means for generating a prosodic parameter editing screen in which a predetermined display symbol is displayed on the graph at the coordinate position specified by the second parameter, and causing the display means to display it,
A program characterized by functioning as The program according to claim 6,
The computer,
In the prosodic parameter editing screen displayed on the display means, a displacement of the coordinate position of the display symbol is received via the input means, and the prosodic parameter of the intermediate language corresponding to the prosodic control unit in which the display symbol is displaced Prosody parameter rewriting means for changing the value of the prosody parameter value to a prosodic parameter value specified from the coordinate position of the display symbol after displacement
A program characterized by further functioning as Prosody parameters including at least information specifying accent, duration length, and fundamental frequency as parameters, and voice data, corresponding to at least one of phoneme display characters and phoneme display characters as prosodic control units A speech synthesizing method in a speech synthesizer comprising a storage means for storing a speech corpus accumulated for each speech control unit and synthesizing speech corresponding to an input character string,
The intermediate language generating means of the speech synthesizer divides the input character string into prosodic control units, and generates an intermediate language in which prosodic parameters are associated with each divided prosodic control unit;
The prosody parameter editing screen generation means of the sound consistency device forms a graph having coordinates in which the values of the first parameter and the second parameter among the parameters included in the prosody parameter are arranged on the horizontal axis and the vertical axis, respectively. Generating a prosodic parameter editing screen in which a predetermined display symbol is displayed on the graph at the coordinate position specified by the first parameter and the second parameter for each of the divided prosodic control units and displaying it on the display means Process,
A speech synthesis method comprising: The speech synthesis method according to claim 8,
The prosody parameter rewriting means of the speech synthesizer accepts the displacement of the coordinate position of the display symbol via the input means on the prosodic parameter editing screen displayed on the display means, and the prosody where the display symbol is displaced Changing the value of the prosodic parameter of the intermediate language corresponding to the control unit to the prosodic parameter value specified from the coordinate position of the display symbol after displacement;
A speech synthesis method, further comprising:

Images