JP2001282274A - Voice synthesizer and its control method, and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To synthesize a natural voice from a phonogram text. SOLUTION: When phonogram text data are inputted from an input device 301, non-sound information is estimated (303). Then a rhythm parameter is generated according to the phonogram text data and non-sound information (304). Then phonemes is selected (305), connected (306), and outputted from an output device 307.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for synthesizing speech from phonetic text, a control method thereof, and a storage medium.

[0002]

2. Description of the Related Art In general, a speech synthesizer for synthesizing Japanese text (FIG. 1) receives a kanji kana mixed sentence and outputs a synthesized waveform. The speech synthesizer is internally
Two processes of a language analysis unit and a sound processing unit are performed. The linguistic analysis unit reads the kanji kana mixed sentence and converts it into text data (herein referred to as phonetic text) representing acoustic information such as accent position and accent strength. FIG. 2 shows an example of the input kanji-kana mixed sentence and the corresponding phonetic text.

[0003] The acoustic processing unit generates a synthesized waveform using the phonetic text.

[0004]

However, in the case of synthesizing speech based on phonogram text as in the above-described speech synthesizer, the speech information included in the phonogram text is considered. However, there is a problem that it is difficult to obtain a natural human voice.

Accordingly, an object of the present invention is to provide a speech synthesizer for synthesizing natural speech from phonetic text, a control method thereof, and a storage medium.

[0006]

In order to solve such a problem, for example, a speech synthesizer according to the present invention has the following arrangement. That is, a speech synthesizer that inputs phonogram text data including acoustic information and performs speech synthesis, and estimating means for estimating non-acoustic information from the input phonogram text;
A parameter setting unit configured to generate a parameter for setting an output form of the synthesized speech based on the non-acoustic information obtained by the estimation unit and the description in the phonetic text data.

[0007]

Embodiments of the present invention will be described below in detail with reference to the accompanying drawings.

FIG. 3 is a functional configuration diagram of the speech synthesis system of the present embodiment.

In FIG. 1, an input device 301 has a keyboard, a mouse, and the like, and is a device for inputting a user's operation to a speech synthesizer 302 which is a center of the present embodiment.

[0010] The speech synthesizer 302 selects a phonetic text corresponding to a predetermined Japanese text in accordance with an operation of the input device 301, and supplies the selected phonetic text to the non-acoustic information estimation module 303. The module 303 includes a word dictionary 30 that describes acoustic information and non-acoustic information of each word.
8 and the connection dictionary 309 describing grammatical connection rules, the non-acoustic information of the input phonogram text is estimated.
This process will be described later in detail. Here, the acoustic information of the word dictionary 308 includes reading information indicating a reading, accent information indicating an accent position and strength, and the like, and the non-acoustic information includes a part of speech information indicating a type of a part of speech and an inflected form.

The module 304 generates various prosodic parameters using the non-acoustic information estimated from the input phonological text and the acoustic information described in the phonological text. This processing will be described later in detail. Module 304
The prosodic parameters obtained by
05.

The phoneme segment selection module 305 selects the most appropriate phoneme segment sequence from the phoneme segment dictionary 310 based on the prosodic parameters obtained in the module 304. The phoneme unit connection module 306 edits and connects each phoneme using the PSOLA method (pitch synchronous waveform superposition method) or the like, and generates one voice waveform data. The audio waveform data thus obtained is output from an output device 307 including a speaker or the like.

FIG. 10 is a specific block diagram of the speech synthesis system of the present embodiment.

In the figure, 1 is a control unit (CPU) for controlling the entire apparatus, 2 is a ROM storing a boot program, a BIOS and the like, and 3 is a program (operation system 10 and the present embodiment) RAM that provides a work area necessary for executing the speech synthesis application program 11). Reference numeral 4 denotes a storage device such as a hard disk device, which is an operating system (OS) 10, the speech synthesis application 11 of the present embodiment, and a phonetic text database 1 that holds phonetic texts corresponding to various Japanese texts.
2. Word dictionary 308, connection dictionary 309, phoneme dictionary 31
0 is stored in advance. The processing units 1 to 4 constitute the above-described speech synthesizer 302.

Reference numeral 5 denotes a keyboard, and 6 denotes a pointing device such as a mouse. The processing units 5 and 6 constitute the input device 301. 7 is a display device. Reference numeral 8 denotes a sound source unit (with a built-in amplifier) for converting the audio waveform data into a sound (also referred to as acoustic) waveform. Reference numeral 9 denotes a speaker that outputs the audio waveform output from the sound source unit 8. Processing unit 7-9
Constitute the output device 307.

In such a configuration, each of the modules 303 to 306 in FIG. 3 is included in the speech synthesis application, and most of them are executed as the processing procedure of the CPU 1. That is, as shown in FIG. 10, the speech synthesis system according to the embodiment requires hardware such as a speaker, but most of the hardware is realized by the processing of the CPU 1. It can be realized with a device.

FIG. 4 is a flowchart showing the overall processing procedure of the speech synthesis application according to this embodiment.

First, when the OS 11 starts up and the speech synthesis application program 12 is loaded into the RAM 3, the process waits until a phonetic text is input (step S401). At this time, the user operates the graphical user interface of the speech synthesis application while watching the display device 7 using the input device 301 such as the keyboard 5 and the mouse 6. The speech synthesis application program 12 responds to an operation input of the input device 301,
A phonetic text corresponding to a predetermined Japanese text is read from the database 12 and input to the speech synthesizer 302.

When the input is confirmed, the module 303
Estimates non-acoustic information from the phonetic text (step S402). The module 304 generates various prosody parameters using the estimated non-acoustic information and the acoustic information included in the phonetic text (step S403). The module 305 uses this parameter and the contents of the phonetic text to select an appropriate phoneme segment from the phoneme segment dictionary 310 (step S404). The module 306 edits and connects them (step S405), and outputs them (step S406).

The above procedure will be described in detail.

First, the processing in step S402 will be described in detail with reference to FIG. In step S402, the word sequence (for example, “synthesis (noun), sound (noun), sound (noun), (Case particle), output (sa-modification), do (verb) ")
Is estimated. Each variable used in the following description is RA
It is assumed that it is secured in M3 in advance.

When a phonetic text including one or more accent phrases is input, the module 303 determines whether or not an unprocessed accent phrase exists (step S5).
01). If no, it is assumed that the processing for all has been completed, and the present processing is terminated.

If there is an unprocessed accent phrase, one is selected and stored in the variable acc (step S502). The accent phrase can be easily extracted by using a space symbol or the like on the phonetic text. Therefore, in the above phonetic text, for example, the first accent phrase “Gothe-on-o” is replaced by the variable acc
Will be stored.

Next, morphological analysis used in kana-kanji conversion and the like is performed on the reading information of the phonogram text of the selected accent phrase, and the word series constituting the phonogram text and the part of speech information of each word are estimated. Yes (Step S50)
3).

Here, when the morphological analysis is performed on the phonetic text reading information, the candidates may not be narrowed down to one (step S504). In this case, module 30
3 performs the process of step S505. On the other hand, as a result of the morphological analysis, when there is no corresponding candidate, the module 303 adds N to the accent phrase stored in the variable acc.
The URL is set, and the processing of the next accent phrase is executed (step S509). This NULL is information indicating that the part of speech information of the word constituting the accent phrase could not be estimated.

The module 303 selects one of the plurality of candidates and stores the selected candidate in a variable kouho (step S505). Next, the module 303 uses the word dictionary 308 and the connection dictionary 309 to change the variable kouho.
Is converted to phonetic text (including acoustic information such as reading information and accent information) and the phonetic text of the candidate is stored in a variable acc2 (step S50).
6).

Next, the phonetic text stored in the variable acc is compared with the phonetic text stored in the variable acc2, and it is determined whether or not the phonetic text matches the accent phrase selected in step S502 (step S507). . If they match, the candidate word series stored in the variable kouho and the part of speech information of each word are set as the word series of the accent phrase and the part of speech information series selected in step S502 (step S508).

For example, as a result of a morphological analysis of the accent phrase “Gorse-on-o”, “work (noun), Kiyone (noun), and (composition particle)” and “synthesis (noun), sound (noun),
Is described below. In this case, the module 303 stores the first candidate “Kyosei Tone” in the variable kouho, and also stores the word candidate 308 in the word dictionary 308 and the connection. Using the dictionary 309, the phonetic text “Go 'Sayon ヲ” of “Kyosei On” is generated and stored in the variable acc2. In this case, the variable ac
Since c and the variable acc2 are not equal, the module 30
3 executes a process for the second candidate.

Next, the module 303 stores the second candidate “synthesized sound” in the variable kouho, and is a phonetic text of “synthesized sound” using the word dictionary 308 and the connection dictionary 309. “Goth-on- ヲ” is generated and stored in the variable acc2. In this case, the variable acc
And the variable acc2 are equal, the module 303
Sets the word sequence of the second candidate and the part-of-speech information of each word to the word sequence and the part-of-speech information sequence of the accent phrase “Goth-on- ヲ”.

These processes are repeatedly executed for all the accent phrases constituting the input phonogram text. After performing a morphological analysis on the phonetic texts of all accent phrases and determining the part of speech information (information indicating the type of part of speech) of each word constituting the phonetic text, the next phonetic text is input. The process ends up to this point.

Next, the processing in step S403 will be described in detail with reference to FIG. In step S403, a process of generating a prosodic parameter of the accent phrase based on the word sequence estimated from the phonetic text of the accent phrase and the part of speech information of each word will be described.

When the phonetic text including one or more accent phrases is input, the module 304 determines whether or not an unprocessed accent phrase exists (step S6).
01). If there is an unprocessed accent phrase, one of them is selected (step S602). Next, one of the unprocessed words constituting the selected accent phrase is selected (step S604). The module 304 calculates prosodic parameters (pitch frequency, phoneme time length, sound power, etc.) of the phoneme sequence corresponding to the selected word based on acoustic information (reading, accent position, accent strength, etc.) corresponding to the selected word. Set to default value (step S60
5).

Next, the module 304 changes the pitch frequency, phoneme time length, and sound power of the phoneme sequence corresponding to the selected word based on the part of speech information that is the non-acoustic information of the selected word (steps S606 to S613). ). here,
A word whose part-of-speech information is NULL is determined as a word for which estimation of part-of-speech information has failed, and all prosodic parameters have default values.

Steps S606 to S613 will be specifically described. When the part of speech of the selected word is determined to be “noun” (step S606), the module 304
The values of the pitch frequency, phoneme time length and sound power are increased (step S610). If the part of speech of the selected word is determined to be a "sentence verb" (step S607), the module 304 increases the values of the pitch frequency, the phoneme duration, and the sound power (step S611). When the part of speech of the selected word is determined to be “numeral” (step S608), the module 304 increases the values of the pitch frequency, the phoneme duration, and the sound power (step S612). When the part of speech of the selected word is determined to be "particle" (step S60)
9) The module 304 decreases the value of the pitch frequency and increases the values of the phoneme duration and the sound power (step S).
613). In the present embodiment, default values are used for parts of speech other than nouns, verbs, numerals, and particles. In the above example, an example in which the prosodic parameter is changed according to the type of part of speech has been described, but the present invention is not limited to this. For example, the prosodic parameters may be changed according to the inflected form of the word.

These processes are performed on all the words constituting all the accent phrases.

As described above, according to the present embodiment,
Prosody parameters of phonetic text can be generated taking into account not only acoustic information but also non-acoustic information (such as the type of part of speech of words), so that more natural speech can be synthesized and output. Become.

<Other Embodiments> In this embodiment, an example will be described in which the prosodic parameters of the entire sentence are changed based on the non-acoustic information of the phonetic text. In the present embodiment, a part of the speech synthesis system of the first embodiment is modified and realized.

When uttering a general sentence, apart from the accent pattern of the accent phrase, the pitch of the accent phrase, the pause phrase, and the sentence (the pitch of the voice) tends to decrease toward the latter half. In the present embodiment, this tendency is defined as a “natural drop component of pitch”. FIG. 7 shows an example of the relationship between the phonetic text, the accent pattern, the natural drop component of the pitch, and the pitch pattern of such a sentence (hereinafter, referred to as a general sentence). In phonetic text,
(Space) ”indicates a delimiter between accent phrases, and“, ”
Indicates a pause phrase, and “.” Indicates the end of a sentence. The pitch pattern is represented by the sum of the accent pattern and the natural drop component of the pitch.

However, when a plurality of nouns are concatenated, such as a place name, when the pitch naturally descends, the utterance becomes natural and comfortable. FIG. 8 shows an example of the relationship between the phonetic text, the accent pattern, the natural drop component of the pitch, and the pitch pattern of such a sentence (hereinafter, referred to as a noun reading sentence).

In the present embodiment, focusing on such features, based on the non-acoustic information obtained from the phonetic text of each accent phrase, whether the sentence composed of the phonetic text is a general sentence, By judging what it is,
A speech synthesis system that generates a synthesized sound with less discomfort will be described.

FIG. 9 is a flowchart for explaining a procedure for generating a prosody parameter in the present embodiment. This processing corresponds to the processing in step S403 of the first embodiment.

Phonetic text composed of one or more accent phrases (here, phonetic text forms one sentence)
Is input, the module 304 checks the part-of-speech information of each accent phrase, and determines whether or not there is a part of speech other than a noun (step S901). Here, if it is determined that the word consists only of a noun, the mode flag is set to 1 (step S903). On the other hand, when there is a part of speech other than a noun, it is regarded as a general sentence, and the mode flag is set to 0 (step S902).

Next, the module 304 determines whether or not an unprocessed accent phrase exists (step S90).
4). If there is an unprocessed accent phrase, one of them is selected (step S905). Next, one of the unprocessed words constituting the selected accent phrase is selected (steps S906 and S907).

Here, in the case of the general sentence mode (mode = 0) (step S908), the module 304 determines the acoustic information (reading, accent position,
Based on the accent strength, etc., the prosodic parameters (pitch frequency, phoneme time length,
Sound power, etc.) are set to default values (step S9)
10). This default value is a value that takes into account the natural drop component of the pitch.

Next, the module 304 changes the pitch frequency, phoneme time length, and sound power of the selected word based on the part of speech information that is non-acoustic information of the word (steps S911 to S918). Here, the part of speech information is NULL
Is determined as a word for which part-of-speech information has failed to be estimated, and all prosodic parameters have default values.

Steps S911 to S918 will be specifically described. When the part of speech of the selected word is determined to be “noun” (step S911), the module 304
The values of the pitch frequency, phoneme time length and sound power are increased (step S915). If the part of speech of the selected word is determined to be a "sentence verb" (step S912), the module 304 increases the values of the pitch frequency, the phoneme duration, and the sound power (step S916). When the part of speech of the selected word is determined to be “numerical” (step S913), the module 304 increases the values of the pitch frequency, the phoneme duration, and the sound power (step S917). When the part of speech of the selected word is determined to be "particle" (step S91)
4), the module 304 decreases the value of the pitch frequency and increases the values of the phoneme duration and the sound power (step S).
918). In the present embodiment, default values are used for parts of speech other than nouns, verbs, numerals, and particles. In the above example, an example in which the prosody parameter is changed according to the type of part of speech has been described, but the present invention is not limited to this. For example, the prosodic parameters may be changed according to the inflected form of the word.

On the other hand, in the case of the noun reading mode (mode = 1) (step S908), the module 304 outputs the acoustic information (reading, accent position,
Based on the accent strength, etc.), the prosodic parameters (pitch frequency, phoneme time length,
Sound power, etc.) are set to default values (step S9)
09). The default value set here is a value that does not consider the natural drop component of the pitch.

With this configuration, "Kyo '
-Ha Gacconi Itta. ), You can set prosodic parameters that take into account the type of part of speech and the natural drop component of pitch,
In a sentence consisting only of nouns such as "Kanakawa 'Ken Kawasaki' Shi Nakahara 'K", it is possible to set a prosodic parameter that does not consider the type of part of speech or the natural drop component of pitch.

In particular, when mode = 0, it is possible to generate a more natural intonation voice, as in the first embodiment described above.

Further, as described above, by controlling the pitch frequency, which is one of the prosody parameters, based on the non-acoustic information (type of speech, etc.), it is possible to reproduce a sound with more intonation. Become. At this time, if the part of speech is an inflection, it is desirable to control the pitch frequency so that the range of the maximum value and the minimum value of the pitch frequency is larger than that of a noun, a numeral, or a particle.

By controlling the sound power, which is one of the prosodic parameters, based on the non-acoustic information in addition to the pitch frequency, a more natural sound can be reproduced. At this time, if the part of speech is a numerical part, it is desirable to control the sound power so that the sound power is larger than that of a noun, an incarnation, or a particle.

By controlling the phoneme time length, which is one of the prosody parameters, based on the non-acoustic information in addition to the pitch frequency and the sound power, a more natural sound can be reproduced. At this time, it is desirable that the control of the phoneme time length is to make the phoneme time length longer than that of a noun, an intransitive verb, or a particle if the part of speech is a numerical part.

In the present embodiment, pitch frequency, sound power, and so on are used as prosodic parameters controlled based on non-acoustic information.
Although the phonological time length has been described, the present invention is not limited to this. For example, the pause time length can be changed based on non-acoustic information. At this time, if the part of speech is a noun or a particle, the pause time length can be controlled by increasing the pause time length immediately after the noun or particle, so that a sound with more natural intonation can be reproduced.

In the embodiment, an example in which the present invention operates on one device has been described. However, a part of the present system may be constructed in a client unit on a wired / wireless network. Anything is fine.

The present invention may be applied to a system constituted by a plurality of devices or to an apparatus constituted by a single device. Further, an object of the present invention is to supply a storage medium (or a recording medium) storing program codes of software for realizing the functions of the above-described embodiments to a system or an apparatus, and to provide a computer (or a CPU or Needless to say, the present invention can also be achieved by an MPU) reading and executing the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention. When the computer executes the readout program codes, not only the functions of the above-described embodiments are realized, but also an operating system (OS) running on the computer based on the instructions of the program codes. It goes without saying that a case where some or all of the actual processing is performed and the functions of the above-described embodiments are realized by the processing is also included.

Further, after the program code read from the storage medium is written into a memory provided in a function expansion card inserted into the computer or a function expansion unit connected to the computer, the program code is read based on the instruction of the program code. , The CPU provided in the function expansion card or the function expansion unit performs part or all of the actual processing,
It goes without saying that a case where the function of the above-described embodiment is realized by the processing is also included.

[0057]

As described above, according to the present invention, speech can be synthesized in consideration of acoustic information and non-acoustic information of phonetic text, so that more natural speech can be output. become.

[Brief description of the drawings]

FIG. 1 is a diagram showing a general processing flow of a speech synthesizer.

FIG. 2 is a diagram showing the relationship between kanji kana mixed sentences and phonetic texts.

FIG. 3 is a functional configuration diagram of the device according to the embodiment.

FIG. 4 is a flowchart illustrating overall processing according to the embodiment.

FIG. 5 is a flowchart of processing for estimating a part-of-speech sequence of an accent phrase in the embodiment.

FIG. 6 is a flowchart of a prosody parameter generation process in the embodiment.

FIG. 7 is a diagram illustrating an outline of control of a pitch frequency of a general sentence.

FIG. 8 is a diagram showing an outline of pitch frequency control in a noun reading mode in the embodiment.

FIG. 9 is a flowchart of a prosody parameter generation process in the embodiment.

FIG. 10 is a block diagram of an apparatus according to the embodiment.

Claims (10)

[Claims] An estimating means for estimating non-acoustic information from phonetic text including acoustic information, and an output form of a synthesized voice is set based on the non-acoustic information obtained by the estimating means and the acoustic information. And a parameter setting means for generating a prosodic parameter for performing the speech synthesis. 2. The speech synthesizer according to claim 1, wherein said estimating means performs an estimating process for each accent phrase constituting said phonetic text. 3. The speech synthesizer according to claim 1, wherein the phonetic text is composed of a symbol indicating a pronunciation and a symbol indicating a position of an accent. 4. The method according to claim 1, wherein said estimating means estimates a word sequence constituting said phonetic text and a part of speech information of each word constituting said word sequence.
The speech synthesizer according to any one of the above items. 5. The prosody parameter is a pitch frequency,
The speech synthesis device according to any one of claims 1 to 4, wherein the speech synthesis device includes at least one of a phoneme time length and power. 6. The parameter setting means controls at least one of a pitch frequency, a phoneme time length, and a power of a phoneme sequence corresponding to the phonetic text according to a part of speech of a word constituting the phonetic text. The speech synthesizer according to any one of claims 1 to 5, wherein: 7. The method according to claim 1, wherein the parameter setting means controls a natural descent component of a phoneme sequence corresponding to the phonetic text according to a part of speech of a word constituting the phonetic text. Item 7. The speech synthesizer according to any one of Items 6 to 6. 8. The parameter setting means, when the part of speech of a word constituting the phonological text is a list of nouns, does not consider a natural descent component of a phonological sequence corresponding to the phonological text. The speech synthesizer according to claim 7, characterized in that: 9. An estimation step of estimating non-acoustic information from phonetic text including acoustic information, and an output form of a synthesized speech is set based on the non-acoustic information and the acoustic information obtained in the estimating step. A parameter setting step of generating a parameter for performing a speech synthesis. 10. A synthesized speech output based on a program code of an estimation step of estimating non-acoustic information from phonetic text including acoustic information, and the non-acoustic information and the acoustic information obtained in the estimating step. A storage medium for storing a program code of a parameter setting step of generating a parameter for setting a form.