JP2001242882A - Method and device for voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice synthesizer which is capable of generating synthesized voice having a variety of prosodical characteristics by text voice synthesis. SOLUTION: A text analysis section 10 conducts a morpheme analysis and a sentence structure analysis for an input text 102 to generate language information 104 (for example, necessary information required for the generation of synthesis parameters such as information on voice symbol trains corresponding to the reading of an input text and on an accent phrase that becomes the unit of cadence control, location o f accents and parts of speech). A synthesis parameter generating section 20 refers to a plurality of cadence control dictionaries 24 to 26 in accordance with the information 104 and generates respectively corresponding synthesis parameters 204 to 206. A synthesis parameter interpolation section 22 conducts interpolation process for cadence parameters of the plural synthesis parameters 204 to 206 in accordance with weighting information 201 to generate synthesis parameters 207. A voice synthesis section 13 generates voice information 108 in accordance with the phoneme information and the cadence information specified by the parameters 207.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesis method and a speech synthesis apparatus for text speech synthesis.

[0002]

2. Description of the Related Art Creating a speech signal artificially from an arbitrary sentence is called text-to-speech synthesis. Usually, a text synthesis system is composed of three stages: text analysis, synthesis parameter generation, and speech synthesis.

FIG. 10 shows a configuration example of a conventional general speech synthesizer. As shown in FIG. 10, a conventional general text synthesizing system generally includes a text analysis unit 10.
10, synthesis parameter generation unit 1001, speech synthesis unit 10
13, a prosody control dictionary 1002. First, the input text 1102 is input to a text analysis unit 1010.
Performs morphological analysis, syntax analysis, and the like, and outputs language information 1104. The linguistic information 1104 includes various information necessary for generating a synthesis parameter, such as a phonetic symbol string corresponding to the reading of the text 1102, information on accent phrases that are units of prosody control, accent positions, parts of speech, and the like. Next, the synthesis parameter generation unit 1001 performs prosody control with reference to the prosody control dictionary 1002 based on the language information 104, and generates the synthesis parameter 1100.
The synthesis parameters 1100 are composed of prosodic parameters such as fundamental frequency, phoneme duration and power, and phoneme parameters such as phoneme symbol strings. Then, the voice synthesizer 10
Reference numeral 13 generates audio information 1108 according to the phonemic information or prosodic information specified by the synthesis parameter 1100.

In such a synthesizing system, it is common to synthesize a voice having a tone (so-called reading tone) as when a human reads out a sentence. In recent years, however, various utterance styles have been controlled by controlling the utterance style. Have been proposed. For example, Japanese Patent Laying-Open No. 10-11083 discloses a method of controlling a speech style of a synthesized speech using a prosody control dictionary of a plurality of speech styles including a reference speech style (reading tone or the like). FIG. 11 shows the configuration of this conventional speech synthesizer.

The difference from the above-described general text synthesis system is that the system has a plurality of prosody control dictionaries (in FIG.
4, 2015), utterance style designation information 2103
Prosody control dictionary (for example, 2015) selected based on
And the reference utterance style prosody control dictionary 2016, respectively.
106, and the utterance style emphasis unit 2012 corrects the utterance style in accordance with the emphasis degree designation information 2101. Plural prosody control information 2014, 2015
Is a prosodic control dictionary of an utterance style different from the reference utterance style, and includes, for example, a conversation style and an announcer style. The utterance style emphasizing unit 2012 calculates the difference between the prosody parameter of the synthesis parameter 2105 of the utterance style selected by the utterance style specification information 2103 and the synthesis parameter 2106 of the reference utterance style, and calculates the emphasis degree specification information 2101 and the difference. By correcting the prosodic parameters of the synthesis parameters 2106 in response, the synthesis parameters 210 whose speech style has been adjusted
7 is generated.

[0006]

As described above, in the conventional speech synthesis method, it is only possible to change the utterance style to an intermediate utterance style between the reference utterance style and the selected one utterance style. The utterance style is constant, and the degree of freedom for change is small. In addition, you can change
Speech style (reading style, conversation style, announcer style, etc.)
There is a problem that it is impossible to change the speaker's individuality (voice of A, voice of B, etc.) and emotion (angry voice, sad voice, etc.).

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has as its object to provide a speech synthesis method and a speech synthesis apparatus capable of improving the variety of prosody of synthesized speech by text speech synthesis. .

[0008]

According to the present invention, a plurality of prosody control dictionaries are used to generate prosody parameters by interpolating the generated prosody parameters at an arbitrary ratio, thereby producing synthesized speech of various prosody. It can be generated. Here, the plurality of prosodic control dictionaries may have different utterance styles, may have different personalities, ages and genders of speakers, or may have different emotions. Those having various characteristics depending on the combination of may be used.

That is, in the speech synthesis method according to the present invention (claim 1), first prosody parameters are respectively generated using a plurality of prosody control dictionaries in accordance with input linguistic information, and According to the designated weight information, interpolation processing is performed between the plurality of first prosody parameters to generate second prosody parameters, and a synthesized speech is generated according to the second prosody parameters.

Here, the linguistic information is obtained by analyzing text such as syllable symbol strings corresponding to reading of text, information on accent phrases which are units of prosodic control, accent positions, parts of speech, hanging parts, and the like. The information obtained,
It is composed of additional information that specifies an average generation speed, a loudness of a voice, and the like.

Here, the prosody control dictionary refers to a prosody such as a fundamental frequency, a phoneme duration, a power, and a pause of a synthesized speech. For example, a typical variation pattern of the fundamental frequency is used. And parameters of a statistical model of a control amount such as an accent component, a phonological duration, a power, and a pause, or a rule represented by a decision tree.

Here, the prosody parameter is a set of parameters that characterize the prosody of the synthesized speech, such as fundamental frequency, phoneme duration, power, and pause.

Here, the interpolation between the prosody parameters is a process of generating an intermediate prosody parameter of the plurality of prosody parameters by a process such as weighted averaging between the plurality of prosody parameters. However, the interpolation process referred to here includes a so-called extrapolation process in which the weight is negative. In this case, the generated prosody parameter is an intermediate value between the plurality of prosody parameters. However, it may be that the characteristics of any of the prosodic parameters are more emphasized. Further, this interpolation processing may be performed for all prosodic parameters, or may be performed for some parameters, for example, only the fundamental frequency. Further, for example, the weight at the time of interpolation may be different between the fundamental frequency and the phoneme duration.

Preferably, the weight information may be changed in a sentence.

Further, in the speech synthesis method according to the present invention (claim 3), a second prosody control dictionary is generated by performing interpolation between a plurality of first prosody control dictionaries, and the inputted linguistic information , A prosody parameter is generated using the second prosody control dictionary, and a synthesized speech is generated according to the prosody parameter.

Here, the interpolating process between the prosodic dictionaries means a phoneme having an intermediate characteristic of the plurality of prosodic control dictionaries by processing such as weighted averaging between corresponding information in the plurality of prosodic control dictionaries. This is a process for generating a control dictionary. However, similar to the above-described interpolation of the prosodic parameters, the interpolation processing here includes a so-called extrapolation processing in which the weight becomes negative. In this case, the generated prosody control dictionary The characteristics may not be intermediate between the plurality of prosody control dictionaries, but may be such that the characteristics of any of the prosody control dictionaries are more emphasized. Further, this interpolation processing may be performed on the entire prosody control dictionary, or may be performed only on a part, for example, only a part related to the fundamental frequency control. Also, for example, the weight for interpolation of the part related to the fundamental frequency and the weight for interpolation of the part related to phoneme duration control may be different.

Further, the speech synthesis method according to the present invention (claim 4) uses a second prosody control dictionary generated by performing interpolation between a plurality of first prosody control dictionaries.
Generate prosodic parameters according to the input linguistic information,
A synthesized speech is generated according to the prosodic parameter.

Preferably, said prosody control dictionary is a representative pattern representing a typical fundamental frequency change pattern;
Alternatively, a pattern having equivalent information, for example, a typical pitch cycle change pattern may be included.

The speech synthesis apparatus according to the present invention (claim 6) includes means for generating first prosody parameters using a plurality of prosody control dictionaries in accordance with input linguistic information, and the prosody control dictionary. Means for performing interpolation between the plurality of first prosody parameters to generate second prosody parameters according to the weight information designated for each, and means for generating synthesized speech in accordance with the second prosody parameters. It is characterized by having.

Further, the voice synthesizing apparatus according to the present invention (claim 7) performs an interpolation process between a plurality of first prosody control dictionaries to generate a second prosody control dictionary, and the input means. A means for generating a prosody parameter using the second prosody control dictionary in accordance with the linguistic information, and means for generating a synthesized speech in accordance with the prosody parameter.

Further, in the method for creating a prosody control dictionary according to the present invention, weight information designated for each of the plurality of prosody control dictionaries is input, and the plurality of first prosody control dictionaries are input according to the input weight information. A second prosody control dictionary is generated by performing an interpolation process.

Further, the prosody control dictionary creation device according to the present invention includes a means for inputting weight information designated for each of the plurality of prosody control dictionaries, and a plurality of first prosody control dictionaries in accordance with the input weight information. Means for generating a second prosody control dictionary by performing an interpolation process between them.

Further, according to the present invention (claim 10), a first prosody parameter is generated using a plurality of prosody control dictionaries in accordance with input linguistic information, and weight information designated for each of the prosody control dictionaries is generated. Computer-readable recording on which is recorded a program for generating a second prosody parameter by performing an interpolation process between the plurality of first prosody parameters according to the following. Medium.

Further, according to the present invention (claim 11), a second prosody control dictionary is generated by performing an interpolation process among a plurality of first prosody control dictionaries, and the second prosody control dictionary is generated in accordance with input language information. Prosody parameters are generated using the prosody control dictionary of
A computer-readable recording medium on which a program for generating a synthesized speech according to the prosodic parameter is recorded.

Note that the present invention relating to the apparatus is also realized as an invention relating to a method, and the present invention relating to a method is also realized as an invention relating to an apparatus.

The present invention relating to an apparatus or a method is provided for causing a computer to execute a procedure corresponding to the present invention (or for causing a computer to function as means corresponding to the present invention, or for causing a computer to correspond to the present invention). The present invention is also realized as a computer-readable recording medium in which a program for realizing the function of performing the above is recorded.

According to the present invention, a prosody parameter is generated by interpolating a prosody parameter generated using a plurality of prosody control dictionaries with an arbitrary weight, and speech synthesis is performed using the prosody parameter. , It is possible to generate synthetic speech having various prosodic features.

According to the present invention, a plurality of prosody control dictionaries are interpolated in advance to generate a prosody control dictionary, and speech synthesis is performed using the prosody control dictionary, thereby increasing the amount of calculation. It is possible to generate synthesized speech having a variety of prosodic features without any problem.

[0029]

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

(First Embodiment) FIG. 1 shows a first embodiment of the present invention.
It is a block diagram showing the example of composition of the speech synthesis device (or speech synthesis software) which realizes the speech synthesis method concerning an embodiment.

As shown in FIG. 1, the speech synthesizer includes a text analysis unit 10, a synthesis parameter generation unit 20,
The apparatus includes a synthesis parameter interpolation unit 22, a speech synthesis unit 13, and a plurality of prosody control dictionaries (three of 24 to 26 in FIG. 1).

The basic configuration and operation of each unit are as follows (basically, processing is performed in the order of the text analysis unit 10, the synthesis parameter generation unit 20, the synthesis parameter interpolation unit 22, and the speech synthesis unit 13). Is done).

The text analysis unit 10 performs morphological analysis and syntax analysis on the input text 102,
The language information 104 is generated. The linguistic information 104 includes various information necessary for generating a synthesis parameter, such as a phonetic symbol string corresponding to the reading of the text 102, information on accent phrases that are units of prosody control, accent positions, and parts of speech.

The synthesizing parameter generation unit 20 outputs the linguistic information 1
In accordance with 04, a synthesis parameter 204 is generated with reference to the prosody control dictionary 24, a synthesis parameter 205 is generated with reference to the prosody control dictionary 25, and a synthesis parameter 206 is generated with reference to the prosody control dictionary 26.

The synthesis parameter interpolating unit 22 calculates the weight information 2
In accordance with 01, the prosody parameters of the synthesis parameter 204, the synthesis parameter 205, and the synthesis parameter 206 are interpolated to generate the synthesis parameter 207.

The voice synthesizing section 13 has a synthesizing parameter 207
Audio information 1 according to the phonetic information or prosodic information specified in
08 is generated.

In the following, an operation example of the present embodiment will be described in detail using a fundamental frequency as an example of a prosody parameter.

Language information 104 is output by the text analysis unit 10.
Is generated, the synthesis parameter generation unit 20 generates a plurality of synthesis parameters by referring to a plurality of prosody control dictionaries in accordance with the language information 104.

FIG. 2 is a functional block diagram showing a process of generating a fundamental frequency in the synthesis parameter generation section 20. FIG. 2 illustrates one prosody control dictionary (actually, the process of FIG. 2 is performed on each of a plurality of prosody control dictionaries).

The fundamental frequency control dictionary 48 is a part of the prosody control dictionary, and includes a representative pattern dictionary 45, a representative pattern selection rule 46, and an offset generation rule 47.
The representative pattern dictionary 45 is a set of typical fundamental frequency change patterns in accent phrase units, and stores, for example, patterns as shown in FIG. The representative pattern selection unit 41 selects a representative pattern 401 expected to be most appropriate from the representative pattern dictionary for each accent phrase, with reference to the representative pattern selection rule 46 according to the language information 104. The offset generation unit 44 generates an offset 404 that specifies the average height of the accent phrase for each accent phrase with reference to the offset generation rule 47 according to the language information 104. The offset processing unit 42 moves the representative pattern dictionary 45 in parallel by the offset 404 on the logarithmic frequency axis, and
02 is generated. The pattern connection unit 43 smoothly connects the fundamental frequency patterns 402 for each accent phrase and outputs a fundamental frequency pattern 403 of the entire sentence. 4A, 4B, and 4C show a representative pattern 401, a fundamental frequency pattern 402, and a fundamental frequency pattern 403, respectively, using the text "Microphone is being tested" as an example.

The synthesis parameter generation unit 20 performs the above-described process of generating the basic frequency pattern by using a plurality of prosody control dictionaries 2.
4, 25, and 26 to generate basic frequency patterns, respectively, and synthesize parameters 204, 205, and 20.
6 are output.

Subsequently, the synthesis parameter interpolation unit 22 performs a weighted average process on the basic frequency patterns of the synthesis parameters 204, 205, and 206 according to the weight information 201 to generate a basic frequency pattern.
07 is output.

The weight information 201 is expressed as a set of n (three in FIG. 1) weighting factors corresponding to n (n is plural) prosody control dictionaries (24 to 26 in FIG. 1). Is done. FIG.
Shows an example of an input unit (for example, by a GUI) of the weight information 201. In this example, the three prosody control dictionaries 24, 25, and 26 correspond to the prosody control dictionaries expressing the prosodic features of the tone of Mr. A, Mr. B, and Mr. C, respectively. At the position of the pointer indicated by a black circle, the prosody to whom and to what extent are designated. The pointer position is
FIG. 6 shows the values of the weight information 201 for a, b, c, and d in FIG. If the position of the pointer is a, weight information 20
1 means that the prosody is intermediate between the three, and in the case of b, B
In the case of c, the prosody most similar to Mr. C also includes some characteristics of Mr. A. In the case of c, the prosody is intermediate between Mr. A and Mr. C, and the characteristics of Mr. B are not included. Is a prosody that exaggerates the features of Mr. A's prosody. As described above, by controlling the weight information, it is possible to generate synthesized speech having various personalities.

As another configuration example of the synthesis parameter interpolation unit 22, it is possible to change the weight information in a sentence. FIG. 7 shows an example of weight information that changes in response to a text corresponding to “this lottery has been won, and you can play and live for a lifetime.” In this example, the three prosody control dictionaries 24, 25, and 26 correspond to prosody of the same person when they are calm, surprised, and glad. By performing the interpolation processing in accordance with such changing weight information, it is possible to generate a fundamental frequency pattern expressing a minute change in emotion.

The weight information 201 may be input by a user, input from another program (process),
Various forms are possible, such as a form provided for each predetermined unit of text (for example, a sentence unit, a constituent element of a sentence), a form generated by analyzing the text by the text analysis unit 10, and the like.

Although the present embodiment has been described using the fundamental frequency control model based on the representative pattern, various other fundamental frequency control models such as a so-called Fujisaki model that approximates the pattern by a function can be used. It can be used.

Further, in the present embodiment, the fundamental frequency has been described as an example of the prosody parameter. However, the prosody parameter such as the prosody duration, power, and pause can be implemented in the same manner. That is, sequences such as phoneme duration, power, and pause are respectively generated using a plurality of prosody control dictionaries, and interpolation processing is performed according to the above-described weight information to synthesize synthesized speech having various prosodic features. Can be generated.

As described above, according to this embodiment, the prosody parameters generated by using a plurality of prosody control dictionaries are interpolated with arbitrary weights to generate the prosody parameters. By performing speech synthesis using the synthesized speech, synthesized speech having various prosody characteristics can be generated.

(Second Embodiment) Next, a second embodiment of the present invention will be described.

In the first embodiment, interpolation processing is performed on the prosody parameters generated with reference to a plurality of prosody control dictionaries. On the other hand, in the second embodiment, a plurality of prosody control dictionaries are obtained. In that a prosody control dictionary in which the characteristics of the prosody are adjusted is created by performing an interpolation process in advance.

FIG. 8 is a block diagram showing a configuration example of a speech synthesizer (or speech synthesis software) for realizing the speech synthesis method according to the present embodiment.

As shown in FIG. 8, the speech synthesizer includes a text analysis unit 10, a synthesis parameter generation unit 30,
The prosody control dictionary interpolation unit 31, the speech synthesis unit 13, and a plurality of prosody control dictionaries (3 to 24 in FIG.
And a prosody control dictionary 32 obtained by interpolation.

The following description focuses on the differences from the first embodiment.

In the present embodiment, the processing is roughly divided into two: a prosody control dictionary interpolation process and a speech synthesis process. That is, the prosody control dictionary 32 is generated in advance by the prosody control dictionary interpolation unit 31 based on the plurality of prosody control dictionaries (24 to 26).
Thereafter, the text analysis unit 1 is used by using the prosody control dictionary 32.
0, synthesis parameter generation unit 30 and speech synthesis unit 13 perform speech synthesis. When the contents of the prosody control dictionary 32 are to be modified or made different, the prosody control dictionary 32 is generated again by the prosody control dictionary interpolation unit 31.

The configuration and operation of the text analysis unit 10, the synthesis parameter generation unit 30, and the speech synthesis unit 13 are basically the same as those in FIG. 10, and the description of each unit is omitted here.

The prosody control dictionary interpolation section 31 will be described below.

The prosody control dictionary interpolation unit 31 performs an interpolation process on the plurality of prosody control dictionaries 24, 25, and 26 according to the weight information 301 to generate a prosody control dictionary 32 (the synthesis parameter generation unit 30 performs , According to the language information 104,
Referring to the prosody control dictionary 32, the synthesis parameters 305
Will be generated).

In the following, an operation example of the present embodiment will be described in detail using a fundamental frequency as an example of a prosody parameter.

The basic frequency control model in the synthesis parameter generation section 30 will be described using a model based on the representative pattern described in FIG. 2, as in the first embodiment.

In this case, the fundamental frequency control dictionary is composed of a representative pattern dictionary, a representative pattern selection rule, and an offset generation rule. However, the representative pattern selection rule is
It is assumed that a prosody control dictionary is created so that all of the plurality of prosody control dictionaries 24, 25, and 26 are common.
In this case, the representative pattern selection rule of the prosody control dictionary 32 may be a copy of any one of the plurality of prosody control dictionaries 24, 25, and 26. Therefore, the processing in the prosody control dictionary interpolation unit 31 includes generating a representative pattern dictionary of the prosody control dictionary 32 by interpolating the representative pattern dictionaries of the plurality of prosody control dictionaries 24, 25, and 26.
The offset generation rules of the prosody control dictionary 32 are generated by interpolating the offset generation rules of the plurality of prosody control dictionaries 24, 25, and 26.

First, the interpolation process of the representative pattern dictionary will be described.

Each representative pattern dictionary is composed of N representative patterns. With respect to the first to N-th representative patterns, the representative patterns having the same number in each representative pattern dictionary are subjected to weighted averaging processing according to the weight information 301, thereby generating an interpolated representative pattern.

Next, the interpolation processing of the offset generation rule will be described.

The offset generation rule can be performed by using quantification class I which is one of the statistical models. FIG. 9 shows an example of an offset generation rule based on quantification type I. The value of the offset is given by the sum of the coefficient corresponding to the category to which each piece of language information belongs and the average value m. For example, a sentence is a sentence position of a accent phrase, the number of moras 4, if the part of speech is a noun, the value of the offset _{becomes m + a 2 + b 4 +} c 1. Therefore, by performing weighted averaging processing on the coefficients corresponding to the same category of each offset generation rule according to the weight information 301, an offset generation rule subjected to interpolation processing is generated.

The weight information 301 is represented by a set of n (three in FIG. 8) weighting factors corresponding to n (n is plural) prosody control dictionaries (24 to 26 in FIG. 8). By generating the prosodic control dictionary by changing the weighting factor,
As in the first embodiment, it is possible to generate synthesized speech having various prosodic features.

Thereafter, speech synthesis can be performed by the text analysis unit 10, the synthesis parameter generation unit 30, and the speech synthesis unit 13 using the prosody control dictionary 32.

It is also possible to prepare a plurality of types of prosody control dictionaries 32 generated with different weight information 301, and to select and use them appropriately.

Although the present embodiment has been described using a fundamental frequency control model based on a representative pattern, various other fundamental frequency control models such as a so-called Fujisaki model that approximates a pattern by a function are used. It is possible.

In the present embodiment, the fundamental frequency control has been described as an example of the prosody parameter. However, the prosody parameter such as the phoneme duration, power, and pause can be implemented in the same manner. That is, the phoneme duration, power, pause, etc.
Since it is possible to control using a statistical model such as a class, the prosody control having various prosodic features is performed by interpolating the model parameters according to the weight information in the same manner as the offset generation rule described above. Information can be generated.

In the present embodiment, the prosody parameters are generated using only the prosody control dictionaries generated by performing interpolation processing on a plurality of prosody control dictionaries in advance. There is an advantage that the amount of calculation is small as compared with the conventional speech synthesis method.

In the present embodiment, the text analysis unit 10 and the synthesis parameter generation unit 3 of the components shown in FIG.
A configuration that is configured as a speech synthesis device (or speech synthesis software) including 0, the speech synthesis unit 13 and the prosody control dictionary 32 is also possible. Alternatively, a speech synthesis device (or speech synthesis software) including the text analysis unit 10, the synthesis parameter generation unit 30, and the speech synthesis unit 13 may be used, and the prosody control dictionary 32 may be separately input. Such a configuration can be used for various purposes, such as creating a prosody control dictionary 32 suitable for the tone of each character of a video game, and using this character to make the character speak in a synthetic voice in the game. It can be applied to a device or an application program.

Similarly, a configuration in which a prosody control dictionary creation device (or a prosody control dictionary creation software) including a plurality of prosody control dictionaries serving as materials for the prosody control dictionary interpolation unit 31 and the prosody control dictionary 32 is also possible. . Alternatively, it may be configured as a prosody control dictionary creation device (or prosody control dictionary creation software) including the prosody control dictionary interpolating unit 31, and a plurality of prosody control dictionaries serving as materials may be separately input. With such a configuration, various prosody control dictionaries 32
Can be created and used by the user himself, or can be created by the manufacturer and provided to the user.

As described above, according to this embodiment, a plurality of prosody control dictionaries are interpolated in advance to generate a prosody control dictionary, and speech synthesis is performed using the prosody control dictionary. Thus, it is possible to generate synthesized speech having various prosodic features without increasing the amount of calculation.

The above functions can be realized as hardware or software.

The present embodiment is also directed to a computer which records a program for causing a computer to execute predetermined means (or for causing a computer to function as predetermined means, or for causing a computer to realize predetermined functions). It can also be implemented as a readable recording medium.

The present invention is not limited to the above-described embodiments, but can be implemented with various modifications within the technical scope thereof.

[0077]

According to the present invention, a prosodic parameter is generated by interpolating a prosodic parameter generated using a plurality of prosodic control dictionaries with an arbitrary weight, and speech synthesis is performed using the prosodic parameter. As a result, it is possible to generate synthesized speech having various prosodic features.

According to the present invention, a plurality of prosody control dictionaries are interpolated in advance to generate a prosody control dictionary, and speech synthesis is performed using the prosody control dictionary, thereby increasing the amount of calculation. It is possible to generate synthesized speech having a variety of prosodic features without any problem.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration example of a speech synthesis device according to a first embodiment of the present invention;

FIG. 2 is a diagram showing a fundamental frequency pattern generation model.

FIG. 3 is a diagram illustrating an example of a representative pattern.

FIG. 4 is a diagram for explaining an example of a fundamental frequency pattern.

FIG. 5 is a diagram for explaining an example of input means for weight information;

FIG. 6 is a diagram showing an example of weight information specified by the input unit of FIG. 5;

FIG. 7 is a diagram for explaining an example of a change in weight information;

FIG. 8 is a diagram showing a configuration example of a speech synthesizer according to a second embodiment of the present invention;

FIG. 9 is a diagram showing an example of an offset generation rule based on quantification class I;

FIG. 10 is a diagram showing a configuration example of a conventional general speech synthesizer.

FIG. 11 is a diagram showing a configuration example of a conventional speech synthesizer for controlling a speech style.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Text analysis part 13 ... Speech synthesis part 20,30 ... Synthesis parameter generation part 22 ... Synthesis parameter interpolation part 24-26,32 ... Prosody control dictionary 31 ... Prosody control dictionary interpolation part 41 ... Representative pattern selection part 42 ... Offset processing unit 43 ... Pattern connection unit 44 ... Offset generation unit 45 ... Representative pattern dictionary 46 ... Representative pattern selection rule 47 ... Offset generation rule 48 ... Basic frequency control dictionary

Claims (9)

[Claims] 1. A first prosody parameter is generated using a plurality of prosody control dictionaries in accordance with input linguistic information, and a plurality of the first prosody is specified in accordance with weight information specified for each of the prosody control dictionaries. A speech synthesis method comprising: performing interpolation processing between parameters to generate a second prosody parameter; and generating a synthesized speech according to the second prosody parameter. 2. The speech synthesis method according to claim 1, wherein said weight information changes in a sentence. 3. A second prosody control dictionary is generated by performing an interpolation process among a plurality of first prosody control dictionaries, and a prosody parameter is generated using the second prosody control dictionary in accordance with input linguistic information. And generating a synthesized speech in accordance with the prosodic parameters. 4. A prosody parameter is generated in accordance with input linguistic information using a second prosody control dictionary generated by performing interpolation between a plurality of first prosody control dictionaries. A speech synthesis method characterized by generating a synthesized speech according to the following. 5. The speech synthesis method according to claim 1, wherein said prosody control dictionary includes a representative pattern representing a typical fundamental frequency change pattern. 6. A means for generating first prosody parameters using a plurality of prosody control dictionaries in accordance with input linguistic information, and a plurality of first prosody parameters in accordance with weight information designated for each of the prosody control dictionaries. A speech synthesis apparatus comprising: means for performing interpolation processing between prosody parameters of the first and second prosody parameters to generate a second prosody parameter; and means for generating a synthesized speech in accordance with the second prosody parameter. 7. A means for performing an interpolation process between a plurality of first prosody control dictionaries to generate a second prosody control dictionary, and using the second prosody control dictionary according to input linguistic information. A speech synthesizer comprising: means for generating a prosody parameter; and means for generating a synthesized speech according to the prosody parameter. 8. A first prosody parameter is generated using a plurality of prosody control dictionaries in accordance with the input linguistic information, and a plurality of the first prosody is specified in accordance with weight information specified for each prosody control dictionary. A computer-readable recording medium storing a program for generating a second prosody parameter by performing an interpolation process between parameters, and generating a synthesized speech in accordance with the second prosody parameter. 9. A second prosody control dictionary is generated by performing an interpolation process between a plurality of first prosody control dictionaries, and a prosody parameter is generated by using the second prosody control dictionary in accordance with input linguistic information. And a computer-readable recording medium recording a program for generating a synthesized speech in accordance with the prosodic parameters.