JP2001134283A - Device and method for synthesizing speech - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize high expressiveness as an alternative conversation means, which has not been realized because it has been difficult to change rhythms and acoustic features in an synthesized speech output. SOLUTION: A language processing part 1 analyzes an input text, and generates language information corresponding to the input text. A phoneme parameter generation means 3 selects a speech synthesis unit as a phoneme parameter corresponding to the input text on the basis of the language information. A rhythm parameter control part 81 generates a rhythm parameter from the language information according to a rhythm parameter generation rule, and also changes the subsequent rhythm parameter according to input operation during speech output. A speech synthesis part 7 synthesizes and outputs speech on the basis of the phoneme parameter and the rhythm parameter.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing apparatus and a voice synthesizing method for synthesizing a voice based on an input text, and more particularly to a prosody or an acoustic method for improving expression when used as a substitute for conversation. The present invention relates to a speech synthesis device and a speech synthesis method whose characteristics can be changed.

[0002]

2. Description of the Related Art When a speech synthesizer is used as a substitute for conversation, there has been a problem that the synthesized speech is monotonous and expressiveness is poor. However, the user changes prosody such as pitch pattern and phoneme duration. A method has been developed to improve the expressiveness of the image. For example, when the user wants to emphasize a certain part in the sentence, the user's intention to emphasize that part can be conveyed to the listener by increasing the accent of the part or making the speech speed easier to hear.

[0003] As a method for such a user to change the prosody of the synthesized speech, a prosody determined according to a predetermined rule is displayed in a graph, and a GUI (Graphical User) is displayed.
There is a method of changing the prosody by a user's operation based on the interface, for example, “Speech creation tool ESSSIGN9 with improved speech expression and production efficiency.
8 "(Mizuno et al., The Acoustical Society of Japan, 1998 Autumn Conference, 2-P-12, pp. 309-310,
(September 1998) (hereinafter referred to as the first document).

FIG. 25 is a block diagram showing a configuration of a first conventional speech synthesizer capable of changing prosody based on the description in the first document. In FIG. 25, reference numeral 101 denotes a language processing unit that analyzes input text to determine accent phrase boundary positions, readings, accent types, and the like, and generates linguistic information including prosody control parameters such as accent strength. , 102 is a speech synthesis unit storage unit that stores speech synthesis units, and 103 reads a speech synthesis unit that matches the reading of the input text from the speech synthesis unit storage unit 102 based on the linguistic information, and generates phoneme parameters. A phonological parameter generation / storage unit 105 that stores the rhythm parameter generation rule for generating a rhythm parameter such as a pitch pattern and a phonological duration from linguistic information. Yes, 106 generates a prosody parameter from linguistic information according to a prosody parameter generation rule A law parameter generating unit, 108 displayed on the screen prosodic parameters as a graph, holds prosodic parameters edited by the user of the GUI operation, the speech synthesizing unit 107 at the time of speech synthesis
A speech synthesis unit 107 generates and outputs synthesized speech based on the prosody parameters and the phoneme parameters.

Next, the operation will be described. Language processing unit 1
01, when input text is supplied, the input text is analyzed to determine accent phrase boundary positions, readings, accent types, etc., and to generate linguistic information including prosody control parameters such as accent strength, etc. Phoneme parameter generation / storage section 103 and prosody parameter generation section 1
06.

[0006] The prosody parameter generation unit 106 reads the prosody control parameters and the reading included in the supplied linguistic information,
In addition, a prosody parameter is determined based on the prosody parameter generation rule in the prosody parameter generation rule storage unit 105 and supplied to the prosody parameter editor 108. The prosody parameter editor 108 displays the supplied prosody parameters as a graph on a screen, holds the prosody parameters edited by the user's GUI operation, and supplies the prosody parameters to the speech synthesis unit 107 during speech synthesis.

On the other hand, phoneme parameter generation / storage section 103
Stores a speech synthesis unit suitable for reading the input text based on the supplied linguistic information in the speech synthesis unit storage unit 102
, And generates a phoneme parameter, holds it until speech synthesis, and supplies it to the speech synthesis unit 107.

At the time of speech synthesis, the speech synthesis unit 107 generates and outputs a synthesized speech based on the supplied prosodic parameters and phonemic parameters.

As described above, the first conventional speech synthesizer generates a synthesized speech after the phonetic parameters and the prosodic parameters are determined by the operation of the user.

As another method for the user to change the prosody of the synthesized speech, there is a method of obtaining conversation information for prosody control from a command embedded in the input text and synthesizing speech based on the conversation information. See, for example,
No. 252784 (hereinafter referred to as a second document).

FIG. 26 is a block diagram showing the configuration of a second conventional speech synthesizer based on the second document. FIG.
, 111 is a command analysis unit that divides an input sentence in which a command for controlling the prosody of the synthesized speech is embedded into text and a command, analyzes the command and generates dialogue information, and 112 denotes a language information. A phoneme parameter generation unit that reads a speech synthesis unit that matches text reading based on the text from the speech synthesis unit storage unit 102 and generates phoneme parameters based on the prosodic control parameter and reading included in the linguistic information. It is a command control prosody parameter generation unit that generates prosody parameters such as pitch patterns and phoneme durations based on prosody parameter generation rules. The other components in FIG. 26 are the same as those in FIG. 25, and thus description thereof will be omitted.

Next, the operation will be described. When supplied with an input sentence in which a command for controlling the prosody of the synthesized speech is supplied, the command analysis unit 111 divides the input sentence into text and a command, analyzes the command, and generates dialog information. The dialog information is supplied to the command control prosody parameter generation unit 113, and the text is supplied to the language processing unit 101 as input text. Language processing unit 1
01 analyzes the input text to generate linguistic information in the same manner as in the first conventional speech synthesizer, and generates the phonological parameter generation unit 112 and the command control prosody parameter generation unit 1.
13.

The phoneme parameter generation unit 112 reads a speech synthesis unit suitable for reading the input text from the speech synthesis unit storage unit 102 based on the supplied linguistic information and supplies the speech synthesis unit 107 as a phoneme parameter.

On the other hand, the command control prosody parameter generation unit 113 generates a pitch pattern based on the prosody control parameters included in the supplied linguistic information, the reading and dialog information, and the prosody parameter generation rules of the prosody parameter generation rule storage unit 105. A prosody parameter such as a sound and phoneme duration is generated and supplied to the speech synthesis unit 107.

The speech synthesizer 107 generates and outputs a synthesized speech based on the prosody parameters and the phoneme parameters.

As described above, in the second conventional speech synthesizer, a phoneme parameter and a prosody parameter are generated in accordance with a command embedded in an input sentence in advance to generate a synthesized speech.

In addition to the above-mentioned technology, Japanese Patent Laid-Open No. Hei 10-11083 discloses a conventional technique for generating a synthesized speech after the phoneme parameter and the prosodic parameter are determined by a user operation.
And those described in Japanese Patent Publication No.

[0018]

Since the conventional speech synthesizer is configured as described above, the first conventional speech synthesizer outputs synthesized speech after the user completes editing of the prosodic parameters. Therefore, it is difficult for the user to change the prosody during voice output, and it is difficult to realize high expressiveness such as emphasizing a part of a sentence according to the response of the other party as a conversation alternative means, Further, in the second conventional speech synthesizer, it is necessary to embed a command in an input sentence in advance in order to change the prosody, so that it is difficult for the user to change the prosody during voice output. There were problems such as difficulty in achieving high expressiveness as a conversation alternative.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and includes a prosody, such as a pitch pattern and a phoneme duration, and a shouting voice, a whispering voice, etc., in response to a user operation during output of a voice. It is an object of the present invention to obtain a speech synthesis device and a speech synthesis method that can be used as a conversational alternative having high expressiveness by changing the voice quality and acoustic characteristics of the speech.

[0020]

A speech synthesizer according to the present invention analyzes an input text and generates linguistic information including at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the input text. A linguistic processing unit, a phonological parameter generation unit that selects a speech synthesis unit corresponding to the input text based on the linguistic information and uses the phonological parameter as a phonological parameter, generates a prosodic parameter from the linguistic information according to the prosodic parameter generation rule, and outputs a voice. It has a prosody parameter control unit that changes the subsequent prosody parameters according to the operation input therein, and a speech synthesis unit that synthesizes and outputs speech based on the phoneme parameters and the prosody parameters.

A speech synthesis apparatus according to the present invention includes a prosody control parameter changing unit for changing a prosody control parameter of linguistic information according to an operation input during voice output, and a prosody parameter from the linguistic information according to a prosody parameter generation rule. And a prosody parameter control unit for generating the prosody parameter generation unit.

A prosody parameter generation rule selection unit for selecting one prosody parameter generation rule according to an operation input during voice output from a plurality of prosody parameter generation rules, and a prosody parameter generation rule selection unit And a prosody parameter control unit for generating a prosody parameter from linguistic information in accordance with the prosody parameter generation rule selected by the unit.

A speech synthesis apparatus according to the present invention includes a prosody parameter generation unit for generating a prosody parameter from linguistic information in accordance with a prosody parameter generation rule, and a first operation for outputting a prosody parameter generated by the prosody parameter generation unit during voice output. And a prosody parameter control unit for changing the prosody parameter according to the input.

In the speech synthesizer according to the present invention, the prosody parameter generation unit generates boundary information having timings of accent phrase boundaries and syllable boundaries based on the prosody parameters and the linguistic information, and the prosody parameter change unit converts the boundary information into boundary information. A part for changing the prosody parameter according to the second operation input during the voice output is selected from the prosody parameter changeable part determined based on the first operation input during the voice output. It changes according to.

In the speech synthesizer according to the present invention, the language processing unit detects a predetermined fixed message part of the input text, analyzes a part other than the fixed message, and analyzes the part other than the fixed message part. Generating linguistic information including at least corresponding accent phrase boundary positions, accent types, reading and prosodic control parameters;
The fixed message selection unit selects the detected fixed message voice waveform data according to the operation input, and the fixed message prosody changing unit changes the voice waveform data according to the operation input during voice output to change the fixed message prosody. Then, the fixed message is reproduced as voice by the voice reproducing unit.

The speech synthesis method according to the present invention analyzes an input text and generates linguistic information including at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the input text;
Selecting a speech synthesis unit corresponding to the input text based on the linguistic information as a phonological parameter, generating a prosodic parameter from the linguistic information in accordance with a prosodic parameter generation rule, and generating a prosodic parameter from the linguistic information according to an operation input during voice output. The method includes a step of changing a prosody parameter, and a step of synthesizing and outputting a speech based on the phoneme parameter and the prosody parameter.

[0027] A speech synthesizer according to the present invention analyzes a input text and generates a linguistic information including at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the input text; A phonological parameter control unit that selects a speech synthesis unit corresponding to the input text based on the linguistic information and sets the phonological parameter as a phonological parameter, and changes a subsequent phonological parameter according to an operation input during voice output, according to a prosodic parameter generation rule. It has a prosody parameter generation unit that generates a prosody parameter from linguistic information, and a speech synthesis unit that synthesizes and outputs speech based on the phoneme parameter and the prosody parameter.

[0028] A speech synthesis unit according to the present invention selects one type of speech synthesis unit set in accordance with an operation input during speech output from a plurality of types of speech synthesis unit sets having different voice qualities and acoustic characteristics. A phoneme parameter control unit configured to select a speech synthesis unit corresponding to the input text based on the linguistic information from the speech synthesis unit set selected by the speech synthesis unit set selection unit and to use the speech synthesis unit as a phoneme parameter; Part.

[0029] In the speech synthesizer according to the present invention, the language processing unit detects a predetermined fixed message portion of the input text, analyzes a portion other than the fixed message, and analyzes the portion other than the fixed message portion. Generating linguistic information including at least corresponding accent phrase boundary positions, accent types, reading and prosodic control parameters;
The fixed message selection unit selects the detected fixed message voice waveform data according to the operation input, and the fixed message prosody changing unit changes the voice waveform data according to the operation input during voice output to change the fixed message prosody. Then, the fixed message is reproduced as voice by the voice reproducing unit.

A speech synthesis method according to the present invention analyzes an input text and generates linguistic information corresponding to the input text, the linguistic information including at least accent phrase boundary position, accent type, reading and prosody control parameters;
A step of selecting a speech synthesis unit corresponding to the input text based on the linguistic information and setting it as a phonological parameter, and changing a subsequent phonological parameter according to an operation input during voice output; The method includes the steps of generating a prosody parameter and synthesizing and outputting speech based on the phoneme parameter and the prosody parameter.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 1 of the present invention, and FIG.
FIG. 3 is a configuration diagram illustrating a configuration example of a language processing unit. In FIG. 1, reference numeral 1 denotes a language processing unit that analyzes input text and generates linguistic information including at least accent phrase boundary positions, accent types, readings, and prosody control parameters corresponding to the input text. In the language processing unit 1 shown in FIG. 2, reference numeral 11 denotes a dictionary storage unit that stores in advance a dictionary in which morpheme notation, reading, part of speech, basic accent type, and the like are described. Reference numeral 12 denotes a dictionary in the dictionary storage unit 11 as appropriate. This is a text analysis unit that determines the accent type and reading of the accent phrase. A prosody rule storage unit 13 stores prosody rules for generating prosody control parameters. As a prosodic rule, for example, the accent strength is calculated using the model of quantification type I according to the part of speech of the independent phrase of the accent phrase and the type of the adjunct word, and the average mora length is calculated by the mora of the accent phrase. It is determined for each number by referring to a table in which the average mora length of the natural speech sample is stored in advance. 14
Determines a prosody control parameter by referring to a predetermined prosody rule based on the accent phrase sequence, and determines an accent phrase boundary position,
It is a prosody control parameter generator that outputs linguistic information along with accent type and reading.

Returning to FIG. 1, reference numeral 2 denotes a speech synthesis unit storage unit for pre-storing a speech synthesis unit as a phoneme parameter (for example, a speech waveform when the basic synthesis system directly uses a speech waveform such as PSOLA). Reference numeral 3 denotes a phoneme parameter generation unit that selects a speech synthesis unit corresponding to the input text based on the linguistic information and uses the selected speech synthesis unit as a phoneme parameter.

Reference numeral 4 denotes a prosody control parameter changing unit for changing a prosody control parameter of the linguistic information according to an operation input during voice output. Reference numeral 5 denotes a prosody parameter generation rule storage unit that stores prosody parameter generation rules for generating prosody parameters such as pitch patterns and phoneme durations from linguistic information. The prosody parameter generation rule is
For example, it includes a model for determining a pitch pattern and a phoneme duration, and a conversion rule for converting a prosody control parameter included in linguistic information into a model parameter.

For example, the prosody parameter generation rules include a point pitch model as a pitch pattern generation model, and a rhythm perception point equally spaced model as a phoneme duration determination model. In the rhythm perception point equally spaced model,
The vowel length is determined by fixing the consonant length and rhythm perception points determined by the time from the consonant-vowel boundary for each syllable type at equal time intervals, and the last vowel length is fixed to the rhythm perception point interval The length is multiplied by a value (for example, 0.80). A rhythm perception point position table and a duration table for each consonant type are stored as the model information. FIG. 3 is a diagram illustrating an example of a rhythm perception point table, and FIG. 4 is a diagram illustrating an example of a consonant duration time table. In FIG. 3, a positive value indicates that the rhythm perception point is temporally later than the consonant-vowel boundary. For a single vowel and a vowel, the phoneme start point is regarded as a consonant-vowel boundary.
The prosody parameter generation rules include, for example, a conversion rule in which the prosody control parameters in the linguistic information are used as model parameters as they are.

Returning to FIG. 1, reference numeral 6 denotes a prosody parameter generation unit for generating a prosody parameter from linguistic information according to a prosody parameter generation rule. The prosody parameter changing unit 4 and the prosody parameter generation unit 6 generate a prosody parameter from the linguistic information in accordance with the prosody parameter generation rule and change a subsequent prosody parameter in response to an operation input during voice output. 81.

Reference numeral 7 denotes a speech synthesizer for synthesizing and outputting speech based on phonemic parameters and prosodic parameters.

Next, the operation will be described. For example, an input text that is a kanji-kana mixed sentence created by a user using a key input or a kana-kanji conversion is used by the language processing unit 1.
Supplied to

The language processing section 1 analyzes the input text and determines the accent phrase boundary position, and also determines prosody control parameters such as accent strength and average mora length in a point pitch model according to predetermined rules. Linguistic information including at least the accent phrase boundary position, the pronunciation, the accent type, and the prosody control parameter are supplied to the phoneme parameter generation unit 3 and the prosody control parameter change unit 4.

At this time, in the language processing unit 1, the text analysis unit 12 appropriately refers to a dictionary in the dictionary storage unit 11 in which, for example, morphological notation, reading, part of speech, basic accent type, etc. are described, and The accent type and pronunciation of the accent phrase are determined, and the prosody control parameter generation unit 14 determines the prosody control parameter by referring to the prosody rules in the prosody rule storage unit 13 based on the accent phrase sequence, and determines the accent phrase boundary position and the accent. Output as linguistic information along with type and reading.

Here, the operation of the language processing section 1 will be described in detail using a specific example. Here, as a specific example, an operation in a case where the input text is a sentence "Tomorrow, the whole family will go to an amusement park." FIG. 5 is a diagram showing an example of an accent phrase string by the text analysis unit, and FIG.
FIG. 4 is a diagram illustrating an example of language information by a language processing unit.

First, the text analysis unit 12 operates in the dictionary storage unit 1.
With reference to the dictionary of No. 1, the accent phrase boundary position ("/" position in the character string) "tomorrow / family / all / amusement park / go" is determined, and the accent type and reading of the accent phrase are determined. 5 is supplied to the prosody control parameter generation unit 14 as an accent phrase string including accent phrases in the order 1 to 5. Next, the prosody control parameter generation unit 14
Refers to prosody rules based on the supplied accent phrase sequence, determines prosody control parameters such as accent strength and average mora length in the point pitch model, and
As shown in (1), linguistic information is output together with the accent phrase boundary position, accent type, pronunciation, and the like. In addition,
Although the accent phrase boundary position is not explicitly described in FIG. 6, the accent type and the like are described for each accent phrase, so that the accent phrase boundary position is substantially described.

In this way, the input text is analyzed by the language processing section 1 and language information is output.

Next, the phoneme parameter generation unit 3 reads out a speech synthesis unit suitable for reading each accent phrase of the input text from the speech synthesis unit storage unit 2 based on the supplied linguistic information, and uses the speech synthesis unit as a phoneme parameter. 7

The prosody control parameter changing section 4 outputs an operation input corresponding to a user's operation on an input means (not shown) such as a lever during voice output (that is, when the voice synthesis of the input text is partially completed). ), The value of the prosody control parameter of the following accent phrase (the next accent phrase after the accent phrase currently being output) included in the supplied language information was changed, and the prosody control parameter was changed. The applied language information is supplied to the prosody parameter generation unit 6. When the operation input is not supplied, the prosody control parameter changing unit 4 supplies the supplied linguistic information to the prosody parameter generation unit 6.

The operation of the prosody control parameter changing unit 4 at this time will be described in detail using a specific example. Here, the operation when the language information shown in FIG. 6 is supplied will be described as a specific example. Further, it is assumed that an operation input is supplied by a user operating an input unit (not shown) having two levers for respectively changing an accent strength and an average mora length of a subsequent accent phrase.

At this time, for example, the lever is operated during the voice output of the second accent phrase "family" in the linguistic information of FIG. 6, and the accent strength is 0.2 octave (oc).
When an operation input instructing to increase by t) is supplied, the accent strength of the subsequent accent phrase "everyone" is changed from 0.3 octave to 0.5 octave.

In this way, the prosody control parameter changing unit 4 supplies the linguistic information from the language processing unit 1 to the prosody parameter generation unit 6 in principle, and if there is an operation input, the prosody control parameter changing unit 4 according to the operation input Applied language information, which is language information with changed control parameters, is supplied.

Next, the prosody parameter generation unit 6 reads the prosody parameter generation rules from the prosody parameter generation rule storage unit 5 and generates the prosody parameters based on the language information or the applied language information from the prosody control parameter change unit 4. That is, when applied linguistic information is supplied, a prosody parameter is generated based on the applied linguistic information.

At this time, the prosody parameter generation unit 6 generates a prosody parameter for each accent phrase. A point pitch model is used for pitch pattern generation, and a point pitch suitable for the number of mora, accent type, and accent strength included in the linguistic information or applied linguistic information is generated. In addition, the phoneme duration is generated using an average mora length included in the linguistic information or the applied linguistic information as a time interval of the rhythm perception point, using a rhythm perception point equidistant arrangement model.

FIG. 7 is a diagram showing an example of a comparison of pitch patterns when the prosody control parameter (accent strength) is changed and when it is not changed. In the linguistic information shown in FIG. 6, there is an operation input for increasing the accent strength by 0.2 octaves during the voice output of the accent phrase "family", and when the applied linguistic information is supplied, the accent phrase "everyone"
As shown in FIG. 7, in the pitch pattern of (reading: miNnade), the accent strength is increased by 0.2 octave. On the other hand, when there is no operation input, the accent strength remains at 0.3 octave as shown in FIG. 7 based on the linguistic information shown in FIG. The vertex of the point pitch pattern is the vowel center point. When an operation input is made, the pitch is increased as shown in the figure, and the accent phrase is emphasized.

FIG. 8 is a diagram showing an example of a comparison of phoneme durations when there is a change in the prosody control parameter (average mora length) and when there is no change. Similarly, in the linguistic information shown in FIG. 6, there is an operation input for increasing the average mora length by 25% during the voice output of the accent phrase "family".
When the applied linguistic information is supplied, as shown in FIG. 8B, the duration of the vowel in the accent phrase "everyone" is increased by 25%, and the accent phrase is emphasized. On the other hand, when there is no operation input, the phoneme duration time becomes as shown in FIG. 8A based on the linguistic information shown in FIG.

Thus, based on the linguistic information or the applied linguistic information, the prosody parameters such as the pitch pattern and the phoneme duration are generated and supplied to the speech synthesizer 7.

The speech synthesizing unit 7 generates the phoneme parameters from the phoneme parameter generation unit 3 and the prosody parameter generation unit 6.
And generates and outputs a synthesized speech according to a predetermined speech synthesis method based on the prosodic parameters from. For example, when a method of directly using a speech waveform such as PSOLA is used, a speech waveform input as a phoneme parameter is subjected to a modification such as cutout or addition superimposition in accordance with the prosodic parameter, and the speech waveforms are directly connected. To generate synthesized speech.

As described above, according to the first embodiment, the prosody control parameters of the linguistic information are changed according to the operation input during the voice output, and the prosody parameters are generated from the linguistic information in accordance with the prosody parameter generation rules. This allows the user to change the prosody, such as the pitch pattern and phoneme duration, in response to the user's operation during the output of the voice, and to partially emphasize the voice during the voice output. Thus, the speech synthesizing apparatus can be used as a conversation alternative having high expressiveness.

Embodiment 2 FIG. 9 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 2 of the present invention. In FIG. 9, reference numeral 21 denotes a multiple prosody parameter generation rule storage unit that stores a plurality of prosody parameter generation rules for generating prosody parameters from linguistic information. Each prosody parameter generation rule has a model and a conversion rule as in the first embodiment. In the second embodiment, the model included in the prosody parameter generation rule is the same for all prosody parameter generation rules. Therefore, in the second embodiment, only the conversion rule differs depending on the prosody parameter generation rule. FIG. 10 is a diagram showing an example of a plurality of types of conversion rules. Note that “−” in FIG. 10 means that the prosody control parameters included in the linguistic information are used as they are as model parameters. As in the first embodiment, a point pitch model is used for generating a pitch pattern, and a consonant is used for generating a phoneme duration as in the first embodiment. A rhythm perception point equally spaced model with a fixed length is used.

Returning to FIG. 9, reference numeral 22 denotes a prosody parameter generation rule selection unit that selects one prosody parameter generation rule from a plurality of prosody parameter generation rules in accordance with a rule selection input (operation input) during voice output. A prosody parameter generation unit 6 generates a prosody parameter from linguistic information in accordance with the prosody parameter generation rule selected by the prosody parameter generation rule selection unit 22.

The prosody parameter generation unit 6 and the prosody parameter generation rule selection unit 22 generate prosody parameters from linguistic information in accordance with the prosody parameter generation rules, and change subsequent prosody parameters according to an operation input during voice output. The prosody parameter control unit 81A is configured.

The other components in FIG. 9 are the same as those according to the first embodiment (FIG. 1), and will not be described.

Next, the operation will be described. Language processing unit 1
The operation of the phoneme parameter generation unit 3 is the same as that of the first embodiment, and a description thereof will be omitted.

The prosody parameter generation rule selection unit 22
When a user's rule selection input is supplied during audio output,
In response to the input, one of the prosody parameter generation rules is read from the plural prosody parameter generation rule storage unit 21 and supplied to the prosody parameter generation unit 6. When the rule selection input is not supplied, a predetermined prosody parameter generation rule is supplied as a default.

For example, the correspondence between the rule numbers and the conversion rule contents shown in FIG. 10 is displayed on the screen, and when the user selects any one of the rule numbers, the result of the selection is used as a rule selection input and the prosody parameter generation rule selection unit. 22.

Then, the prosody parameter generation unit 6 generates prosody parameters based on the linguistic information from the language processing unit 1 in accordance with the prosody parameter generation rules supplied by the prosody parameter generation rule selection unit 22. For example, when the linguistic information shown in FIG. 6 is supplied to the prosody parameter generation unit 6 and the rule selection input of the rule number 2 is supplied to the prosody parameter generation rule selection unit 22 during the audio output of the accent phrase “family”, The accent strength of the accent phrase "everyone" is increased by 0.2 octaves, and the same pitch pattern and phoneme duration as in the case with the operation input in FIG. 7 are generated.

The speech synthesizer 7 synthesizes and outputs speech based on the phoneme parameters from the phoneme parameter generator 3 and the prosody parameters from the prosody parameter generator 6 in the same manner as in the first embodiment.

As described above, according to the second embodiment, one prosody parameter generation rule is selected from a plurality of prosody parameter generation rules in accordance with an operation input during voice output, and according to the selected prosody parameter generation rule. Since prosodic parameters are generated from linguistic information, the prosody such as pitch pattern and phoneme duration is changed according to the user's operation during speech output, and partial emphasis of speech during speech output is performed. An effect is obtained that the speech synthesizer can be used by the user and can be used as a conversation alternative having high expressiveness.

Embodiment 3 FIG. 11 is a block diagram showing a configuration of a voice synthesizing apparatus according to Embodiment 3 of the present invention. In FIG. 11, reference numeral 31 denotes a prosody parameter changing unit that changes a prosody parameter according to a prosody change operation input (first operation input) during voice output. The prosody parameter generation unit 6 and the prosody parameter change unit 31 generate a prosody parameter from the linguistic information in accordance with the prosody parameter generation rule and change a subsequent prosody parameter according to an operation input during voice output.
1B.

The other components in FIG. 11 are the same as those according to the first or second embodiment (FIGS. 1 and 9), and a description thereof will be omitted.

Next, the operation will be described. Language processing unit 1
The operation of the phoneme parameter generation unit 3 is the same as that of the first embodiment, and a description thereof will be omitted.

The prosody parameter generation unit 6 reads the prosody parameter generation rules from the prosody parameter generation rule storage unit 5, generates the prosody parameters based on the linguistic information from the language processing unit 1, and supplies them to the prosody parameter change unit 31. . For example, when the linguistic information shown in FIG. 6 is supplied to the prosody parameter generation unit 6, the same pitch pattern and phoneme duration as those in the case where there is no operation input in FIG. You.

Next, when the prosody parameter changing unit 31 is supplied with the user's prosody changing operation input during the voice output, the prosody parameter generation unit 6 according to the operation input only while the operation input is being supplied. Is changed, and the changed prosody parameter is supplied to the speech synthesizer 7.

The operation of the prosody parameter changing unit 31 at this time will be described in detail using a specific example. The prosody change operation input in the third embodiment is, for example, 2 for changing the pitch of the subsequent voice and the phoneme duration.
It is assumed that the information is supplied in response to a user's operation performed on an input unit (not shown) including the lever. FIG.
FIG. 8 is a diagram showing an example of a pitch pattern changed according to a prosody changing operation input.

In this case, while the corresponding lever is operated by the user, the operation amount (for example, the angle of the lever) is supplied as the prosody change operation input, and the pitch of the voice and the phoneme duration are added to the operation amount. It is changed by the corresponding ratio.

For example, it is assumed that the change of the phoneme duration is effective only at the time of outputting a vowel, and the linguistic information shown in FIG. 6 is supplied to the prosody parameter generation unit 6. When the user operates the lever for changing the pitch so as to raise the pitch during the voice output corresponding to the phoneme sequence “miNna” to emphasize “everyone” of “ As shown in FIG. 12, the pitch pattern is changed. Also, in this case, in order to lengthen the time length of “mi” and emphasize “everyone”, when the user operates the lever so that the time length doubles during the voice output of the syllable “mi”, for example. , The phoneme duration of the vowel / i / is changed to 300 milliseconds in response to the operation.

The speech synthesizing unit 7 converts the phoneme parameters from the phoneme parameter generation unit 3 and the prosody parameter changing unit 3
The speech is synthesized based on the prosodic parameter from No. 1 and output.

As described above, according to the third embodiment, the prosody parameters generated by the prosody parameter generation unit 6 are changed in accordance with the prosody change operation input during voice output. The effect is obtained that the user can instantly perform partial emphasis on the synthesized speech and the like, and that the speech synthesis apparatus can be used as a conversation alternative having high expressiveness.

Embodiment 4 FIG. 13 is a block diagram showing a configuration of a voice synthesizing apparatus according to Embodiment 4 of the present invention. In FIG. 13, 6A generates prosody parameters from linguistic information in accordance with a prosody parameter generation rule, and generates boundary information having timings of accent phrase boundaries and syllable boundaries based on the generated prosody parameters and linguistic information. A prosody parameter generation unit (prosody parameter generation unit) 31A includes a prosody parameter change unit that determines the prosody from a changeable portion of the prosody parameter determined based on the boundary information according to a prosody change section input (second operation input) during voice output. A part whose parameter is to be changed is selected, and the prosody parameter of that part is input to the prosody changing operation input (first
Is a section-specific prosody parameter changing unit (prosody parameter changing unit) that changes in accordance with the operation input of the above. The boundary information / prosodic parameter generation unit 6A and the section-specific prosody parameter changing unit 31A generate prosody parameters from linguistic information in accordance with the prosody parameter generation rules, and change subsequent prosody parameters according to an operation input during voice output. It constitutes the prosody parameter control unit 81C.

The other components in FIG. 13 are the same as those according to the third embodiment (FIG. 11), and will not be described.

Next, the operation will be described. Language processing unit 1
The operation of the phoneme parameter generation unit 3 is the same as that of the first and third embodiments, and the description thereof is omitted.

The boundary information / prosodic parameter generation unit 6A
The prosody parameter generation rule is read from the prosody parameter generation rule storage unit 5, a prosody parameter is generated based on the linguistic information from the linguistic processing unit 1, and based on the linguistic information and the phoneme duration of the generated prosody parameter. , Generates boundary information having the time of the accent phrase boundary and the time of the syllable boundary, and supplies the prosody parameter and the boundary information to the section-designated prosody parameter changing unit 31A.

For example, when the linguistic information shown in FIG. 6 is supplied to the boundary information / prosodic parameter generation unit 6A, the same pitch pattern and phoneme as in the case of no operation input in FIG. A duration is generated. FIG. 14 is a diagram showing an example of the boundary information generated by the boundary information / prosodic parameter generation unit 6A based on the linguistic information shown in FIG. FIG. 14A shows an example of the accent phrase boundary time at this time, and FIG.
FIG. 4B shows an example of the syllable boundary time of the accent phrase “everyone”.

Then, the section designation prosody parameter changing unit 31
A presents a candidate of a section in which the prosodic parameter can be changed to the user based on the supplied boundary information, and determines a section in which the prosodic parameter is changed based on the input of the prosody changing section by the user during voice output. Then, based on the input of the prosody changing operation by the user who is outputting the voice, the prosody parameter of the section is changed and supplied to the voice synthesizer 7. In addition,
When either the prosody change operation input or the prosody change section input is not supplied, the section designation prosody parameter change unit 31A supplies the prosody parameter from the boundary information / prosody parameter generation unit 6A to the speech synthesis unit 7 as it is.

Here, the section designation prosody parameter changing unit 31
The operation of A will be described in detail using a specific example. The prosody change section input in the fourth embodiment is performed by, for example, touching an accent phrase or syllable displayed on a screen of a display input device (not shown) such as a display and a touch panel disposed on the screen. Accent phrases or syllables are supplied as prosodic change interval inputs. In addition, the prosody change operation input in the fourth embodiment is performed, for example, to an input unit (not shown) having two levers for changing the voice pitch and the phoneme duration of the designated section, respectively. It shall be supplied in response to a user operation.

FIG. 15 is a diagram showing an example of a display screen of a display / input device for performing section input. As shown in FIG. 15, syllables constituting the text being output as speech are displayed for each accent phrase on the screen, and a cursor is displayed at the syllable being output. For example, designation of a single syllable by pressing one syllable display portion, designation of an entire accent phrase by pressing a “*” display portion at the beginning of a line, designation of a syllable string by pressing two syllable display portions A prosody change section is designated by the above method, and is supplied as a prosody change section input.

FIG. 16 is a diagram showing an example of designation of a prosody change section in the display shown in FIG. As shown in FIG. 16, when two syllables “mi” and “na” are pressed, the syllable string “everyone” is the target whose prosody is changed based on the prosody changing operation input. In the fourth embodiment, the pitch or phoneme duration of the voice in the specified section is changed by a ratio corresponding to a constant value (for example, the maximum value of the lever angle change) specified by the prosody changing operation input. .

For example, in the above case, assuming that the change in the phoneme duration is valid only for the vowel and the vowel sound, the lever corresponding to the change in the phoneme duration is operated so as to increase the time by 50%. Sometimes
/ I /, / N /, and / a included in the section "Everybody"
The time length of / is changed. FIG. 17 is a diagram illustrating an example of the phoneme duration of the syllable string “Minna ga” when the operation is performed to increase the time length by 50% and when the operation is not performed. When there is no prosody change section input, FIG.
As shown in (a), the prosody change operation input is invalidated, the prosody parameter generated by the boundary information / prosody parameter generation unit 6A is supplied to the speech synthesis unit 7 as it is, and when there is a prosody change section input, the prosody change is performed. The phoneme duration is changed as shown in FIG. 17B in response to the change operation input, and the changed prosody parameters are supplied to the speech synthesizer 7.

The speech synthesis unit 7 synthesizes and outputs speech based on the phoneme parameters from the phoneme parameter generation unit 3 and the prosody parameters from the section designation prosody parameter change unit 31A in the same manner as in the first embodiment.

As described above, according to the fourth embodiment, boundary information having timings of accent phrase boundaries and syllable boundaries is generated based on the generated prosody parameters and linguistic information, and based on the boundary information. Select a part to change the prosody parameter according to the input of the prosody change section in the audio output from the determined prosody parameter changeable part, and change the prosody parameter of that part in accordance with the prosody change operation input in the audio output Since the pitch pattern, the prosody such as the phoneme duration is changed according to the user's operation during the output of the voice, the user can partially emphasize the voice during the voice output, and the like. An effect is obtained that the speech synthesizer can be used as a conversation alternative having high expressiveness.

Embodiment 5 FIG. 18 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 5 of the present invention. In FIG. 18, reference numeral 41 denotes a speech synthesis unit set storage unit that stores a plurality of speech synthesis units having different voice qualities and acoustic characteristics, such as a normal voice, a roaring voice, and a whispering voice, as a voice synthesis unit set. It is assumed that the speech synthesis unit set according to the fifth embodiment has a number and a comment indicating an acoustic feature.

Reference numeral 42 designates 1 according to an audio feature selection input (operation input) during voice output from a plurality of types of voice synthesis unit sets.
A speech synthesis unit set selection unit for selecting a type of speech synthesis unit set. Reference numeral 3 denotes a speech synthesis unit corresponding to the input text based on the linguistic information from the speech synthesis unit set selected by the speech synthesis unit set selection unit. A phoneme parameter generation unit that selects and sets phoneme parameters. Note that the phoneme parameter generation unit 3 and the speech synthesis unit set selection unit 42 select a speech synthesis unit corresponding to the input text based on the linguistic information, set the speech synthesis unit as a phoneme parameter, and perform subsequent speech synthesis in accordance with an operation input during speech output. The phoneme parameter control unit 82 that changes phoneme parameters is configured.

The other components in FIG. 18 are the same as those according to the third embodiment (FIG. 11), and will not be described.

Next, the operation will be described. Language processing unit 1
The operation of the prosody parameter generation unit 6 is the same as that of the third embodiment, and the prosody parameter generated by the prosody parameter generation unit 6 is supplied to the speech synthesis unit 7.

Upon receiving the audio feature selection input by the user who is outputting the voice, the voice synthesis unit set selection section 42
The speech synthesis unit set is read out from the speech synthesis unit set storage unit 41 in accordance with the sound feature selection input and held.
On the other hand, when the audio feature selection input by the user is not supplied, the default speech synthesis unit set selection unit 42 selects the default speech synthesis unit set. As described later, for example, by presenting the user with a list of speech synthesis unit sets and a list of comments on the screen, and operating the predetermined input means (not shown) to select the number,
The sound feature selection input is supplied to the speech synthesis unit set selection unit 42.

Here, the operation of the speech synthesis unit set selection section 42 will be described in detail using a specific example. FIG. 19 is a diagram showing an example of a display by the speech synthesis unit set selection unit 42. For example, when a display and a touch panel (the above input means) arranged on the screen are used as a display input device, and the user presses a number and a comment displayed at the bottom of the screen as shown in FIG. An audio feature selection input for instructing selection of a speech synthesis unit set to be performed is supplied to the speech synthesis unit set selection unit 42. The default of the speech synthesis unit set is "normal voice".

While the voice output is being performed, the cursor sequentially moves to the syllable being output as shown in FIG. For example, by pressing the display portion of "3: whisper" immediately before the cursor moves to the syllable "Y" and releasing it when the cursor moves to the syllable "Ni", a voice synthesis unit set corresponding to "3: whisper" Is input to the speech synthesizing unit set selecting unit 42, and the part "Yuenchi ni" is speech-synthesized with a whisper.

At this time, the speech synthesis unit set selection section 42
Selects the whispering voice synthesis unit set while the display portion of "3: Whispering voice" is pressed, and otherwise selects and holds the normal voice voice synthesis unit set.

Then, based on the linguistic information from the linguistic processing unit 1, the phoneme parameter generating unit 3 reads the input text included in the linguistic information from the utterance synthesizing unit set held in the utterance synthesizing unit set selecting unit. A suitable speech synthesis unit is selected and supplied to the speech synthesis unit 7 as a phoneme parameter.

The speech synthesizer 7 synthesizes and outputs speech based on the phoneme parameters from the phoneme parameter generator 3 and the prosody parameters from the prosody parameter generator 6 in the same manner as in the first embodiment.

As described above, according to the fifth embodiment, one set of a plurality of types of speech synthesis units having different voice qualities and acoustic features is provided in accordance with the acoustic feature selection input being outputted.
The type of speech synthesis unit set is selected, and the speech synthesis unit corresponding to the input text is selected from the selected speech synthesis unit set based on the linguistic information and used as the phonological parameter. By changing the sound characteristics and the acoustic characteristics, the user can instantly perform partial emphasis on the synthesized speech being output, and can use the present speech synthesis apparatus as a conversation alternative having high expressiveness. The effect is obtained.

Embodiment 6 FIG. FIG. 20 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 6 of the present invention, and FIG. 21 is a block diagram showing a configuration example of a fixed message determination language processing unit in FIG. In FIG. 20, 51
Detects a predetermined fixed message part of the input text, analyzes a part other than the fixed message, and includes at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the part other than the fixed message. A fixed message determination language processing unit (language processing unit) that generates language information.

In the fixed message determination language processing unit 51 shown in FIG. 21, reference numeral 61 denotes a fixed message list storage unit for storing texts of fixed messages with numbers.
In the sixth embodiment, for example, "Hello",
A frequently used fixed phrase such as "Good morning" or "How much is it?" Is stored as a fixed message. 62 detects a portion of the input text that matches one of the fixed messages stored in the fixed message list storage unit 61, supplies the number of the detected fixed message to the fixed message selection unit 53, and This is a fixed message determination unit that supplies other parts to the text analysis unit 12.

Returning to FIG. 20, reference numeral 52 denotes a fixed message storage unit for storing a plurality of fixed message data, which is a set of a fixed message number and a plurality of types of voice waveform data corresponding thereto. FIG. 22 shows an example of fixed message data. As shown in FIG.
Has a voice waveform data number and a comment for selecting one voice waveform data by the fixed message selection unit 53 in response to a message selection input by the user, and a voice file. In addition,
In the fixed message data shown in FIG. 22, the audio file is specified by the file name. Further, in the sixth embodiment, since a method inferred from PSOLA is used as a method for changing the prosody of a fixed message as described later, the audio file has an audio waveform and information on a pitch synchronization point.

Reference numeral 53 denotes a fixed message selection unit for selecting voice waveform data of a fixed message according to a fixed message number and a message selection input (operation input).
Reference numeral 4 denotes a fixed message prosody changing unit which changes the fixed message prosody by changing the voice waveform data of the fixed message in response to the fixed message prosody changing operation input (operation input) during voice output. This is an audio playback unit that plays back as a sound.

The other components in FIGS. 20 and 21 are the same as those according to the first embodiment (FIGS. 1 and 2), and a description thereof will be omitted.

Next, the operation will be described. When the input text is supplied, the fixed message determination language processing unit 51 converts the input text by the fixed message determination unit 62 into a portion that matches the fixed message stored in the fixed message list storage unit 61 in advance and a portion other than the fixed message. The fixed message number of the matched fixed message is supplied to the fixed message selection unit 53, and the text analysis unit 1
2, the text other than the fixed message is analyzed to determine the accent phrase boundary position, and the prosody control parameter generation unit 14 determines the prosody control parameters such as accent strength and average mora length according to predetermined rules. It outputs linguistic information including at least reading, accent type and prosody control parameters.

The operation of the fixed message determination language processing section 51 will be described in detail using a specific example. For example, an input text that is a kanji-kana mixed sentence created by a user using a key input or a kana-kanji conversion is supplied to the fixed message determination language processing unit 51. Here, as a specific example, an operation in a case where the input text is a sentence “Good morning. The whole family will go to an amusement park tomorrow.”

The fixed message judging section 62 supplied with the input text refers to the fixed message list in the fixed message list storage section 61 and determines the input text as a fixed message "Good morning" and a message other than the fixed message. The text is divided into "Tomorrow, the whole family will go to the amusement park". The fixed message number of the fixed message "Good morning" is supplied to the fixed message selection unit 53, and the text other than the fixed message is analyzed by the text analysis unit. 12

The operation of text analysis unit 12 and prosody control parameter generation unit 14 to which the text is input is the same as the operation of text analysis unit 12 and prosody control parameter generation unit 14 of language processing unit 1 in the first embodiment. Therefore, the description is omitted. Also, the operations of the phoneme parameter generation unit 3, the prosody control parameter change unit 4, the prosody parameter generation unit 6, and the speech synthesis unit 7 are the same as those in the first embodiment, and thus the description thereof is omitted.

On the other hand, the fixed message selecting section 53 reads out the audio waveform data corresponding to the supplied fixed message number from the fixed message storage section 52, and displays, for example, the number of the audio waveform data and the comment on the screen, and responds to it. Speech waveform data of a type of voice quality according to the user's message selection input is selected and supplied to the fixed message prosody changing unit 54 as a reference fixed message. For example, in the case of the fixed message data shown in FIG. 22, when the fixed message "Good morning" is detected, 2 is supplied as the fixed message number.

FIG. 23 is a diagram showing an example of voice waveform data numbers and comments displayed on the screen corresponding to the fixed message data of FIG. For example, when these are displayed on a display unit (display) of a display and a display input device such as a touch panel arranged on the screen, and a user presses, for example, a portion of number 3 on the input unit (touch panel), the message selection input is performed. The number 3 is supplied to the fixed message selection unit 53, and the fixed message selection unit 53 uses the audio waveform data of the audio file with the file name “d0203” from the fixed message data as the fixed message prosody change unit 5
4

The fixed message prosody changing unit 54 changes the voice waveform data of the reference fixed message based on the fixed message prosody changing operation input by the user who is outputting the voice, and supplies the changed voice waveform data to the voice reproducing unit 55. I do.

Here, the operation of the fixed message prosody changing unit will be described in detail using a specific example. The fixed message prosody changing operation input in the sixth embodiment is performed by, for example, a user's input performed on input means (not shown) having two levers for respectively changing the pitch of the subsequent voice and the phoneme duration. It shall be supplied in response to the operation. Then, the operation amount (for example, the angle of the lever) is supplied as a fixed message prosody changing operation input, and the pitch of the voice and the phoneme duration are changed by a ratio corresponding to the operation amount. Further, in the sixth embodiment, it is assumed that the change of the phoneme duration is effective only when a vowel is output.

In the fixed message prosody changing unit 54, when the lever is operated to increase the pitch,
That is, when shortening the pitch cycle, the original speech waveform is cut out using a Hanning window twice as long as the changed pitch cycle around the pitch synchronization point, and is added and superimposed at intervals of the changed pitch cycle. On the other hand, when the lever is operated to lower the pitch, that is, when the pitch period is lengthened, the window is cut out with a window length twice as long as the original voice waveform pitch period,
Similarly, superposition is performed. In this way, the pitch of the fixed message is changed.

When the lever is operated to increase the phoneme duration, the fixed message prosody changing section 54 repeats the waveform for one cycle to obtain the designated phoneme duration. On the other hand, when the lever is operated so as to shorten the phoneme duration, the waveform is thinned out in pitch cycle units. Thus, the phoneme duration of the fixed message is changed.

The voice reproducing unit 55 outputs the fixed message as the reproduced voice based on the voice waveform data from the fixed message prosody changing unit 54.

As described above, according to the sixth embodiment, the text other than the fixed message in the input text is subjected to voice synthesis in the same manner as in the first embodiment, and the fixed message in the input text is output during voice output. Select the voice waveform data of the fixed message of voice quality according to the operation input of, change the voice waveform data according to the operation input during voice output, change the prosody of the fixed message,
Since the fixed message is played as audio,
Regardless of whether the message is a fixed message or other text, the user can perform partial emphasis on the synthesized speech being output, and can use the speech synthesis apparatus as a conversational alternative having high expressiveness. The effect that it can be obtained is obtained.

The speech synthesizing apparatuses according to the second to fifth embodiments may further include the prosody control parameter changing unit 4 according to the first embodiment.

Further, in the speech synthesizing apparatus according to the third to fifth embodiments, a plurality of prosody parameter generation rule storage units 21 according to the second embodiment are provided instead of the prosody parameter generation rule storage unit 5, and the prosody parameter according to the second embodiment is provided. The generation rule selection unit 22 may be further provided.

Further, the voice synthesizing apparatus according to the fifth embodiment may further include the prosody parameter changing unit 31 according to the third embodiment.

Further, in the speech synthesizers according to the second to fifth embodiments, the fixed message determination language processing unit 51 according to the sixth embodiment is provided in place of the language processing unit 1, and the fixed message storage unit 52 according to the sixth embodiment is used. A fixed message selecting unit 53, a fixed message prosody changing unit 54, and a voice reproducing unit 55 may be further provided.

Further, the prosodic parameter generation rules in the first to sixth embodiments include a model other than the point pitch model, such as the Fujisaki model, as the pitch pattern generation model. Fixed message determination language processing unit 5 according to Embodiment 6
The linguistic information generated by 1 includes parameters of another pitch pattern generation model to be used (for example, the magnitude of the accent command in the case of the Fujisaki model) instead of the accent strength in the point pitch model as a prosody control parameter. It may be included.

Further, in the prosodic parameter generation rules in the first to sixth embodiments, a model in which a vowel-consonant boundary is arranged at an interval of an average mora length, a duration of a vowel and a consonant, as a phoneme duration generation model. A rhythm perception point, such as a model in which the vowel centroid point (the point at which the integrated value of the power from the vowel start point becomes half the integrated value of the power of the entire vowel) is placed at intervals of the average mora length while maintaining the ratio. Models other than the interval arrangement model may be included.

Furthermore, the linguistic information generated by the language processing unit 1 in the first to fourth embodiments or the fixed message determination language processing unit 51 in the sixth embodiment includes accent strength and average mora length as prosody control parameters. In addition, the average power may be included. In that case, in the first, third, and fourth embodiments, the prosody parameter generation rule includes a model for generating a time-varying pattern of power based on the average power. The parameter changing unit 4, the prosody parameter changing unit 31 in the third embodiment, or the section-specific prosody parameter changing unit 31A in the fourth embodiment,
A prosody control parameter or a prosody parameter is changed according to an operation input or a prosody change operation input by a user to change the power of the synthesized speech. In this case, in the second embodiment, the prosody parameter generation rule includes a model for generating a time change pattern of the average power, and the conversion rule included in the prosody parameter generation rule applies the average power to the model. Include a way to do it. FIG. 24 is a diagram illustrating an example of a conversion rule when the average power is applied to the model.

Further, the amount of parameter change due to the operation input in the first and sixth embodiments, the prosody change operation input in the third and fourth embodiments, and the fixed message prosody change operation input in the sixth embodiment is obtained by replacing the lever angle with the lever angle. Then, the amount of rotation of the dial as the input means, the pressing time of the button as the input means, the angle of the joystick as the input means, determined according to the continuous amount such as the pressure applied to the pressure sensor as the input means Alternatively, a discrete value of the amount of change may be prepared in advance and displayed on the screen, and determined according to a value selected by key input such as a cursor key or a return key, mouse operation, or pressing on a touch panel. It may be.

Further, the rule selection input in the second embodiment, the prosody change section input in the fourth embodiment, and the phonological feature selection input in the fifth embodiment are performed by pressing numbers or syllables displayed on a touch panel or the like. Alternatively, the information may be supplied in response to a key input such as a cursor key or a return key, or a mouse operation.

Furthermore, instead of invalidating the prosody changing operation input when there is no prosody changing section input, the section specifying prosody parameter changing section 31A in the fourth embodiment is similar to the prosody parameter changing section 31 in the third embodiment. Then, the prosody parameter of the subsequent voice may be changed.

Further, in the input text in the first to sixth embodiments, instead of a kanji-kana mixed sentence created by the user based on key input and kana-kanji conversion, a plurality of sentences created and registered in advance by the user or a third party are included. A sentence mixed with kanji or kana characters selected by the user from the sentence candidates may be used.

Further, in the input text in the first to sixth embodiments, instead of a kanji-kana mixed sentence created by a user based on key input and kana-kanji conversion, an accent created and registered in advance by a user or a third party is used. A kanji-kana sentence with linguistic information selected by the user from a plurality of sentence candidates with linguistic information such as phrase boundaries, accent types, parts of speech, and mora number may be used. In this case, the text analysis processing in the language processing unit 1 and the fixed message determination language processing unit 51 is unnecessary.

Furthermore, the input text in the first to sixth embodiments is a sentence composed of a variable part and a fixed part in which words are embedded, instead of a kanji-kana mixed sentence created by the user based on key input and kana-kanji conversion. A fill-in-the-blank sentence in which the user inputs a word by key input or candidate selection may be used for the template. In the case where the fixed message in Embodiment 6 uses the padding sentence as the input text, the fixed message may include a fixed part of the sentence template in addition to the frequently used sentence itself.

[0128]

As described above, according to the present invention, an input text is analyzed, and linguistic information including at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the input text is generated. Based on the linguistic information, a speech synthesis unit corresponding to the input text is selected as a phonological parameter, and a prosodic parameter is generated from the linguistic information according to a prosodic parameter generation rule, and a subsequent prosody parameter is generated according to an operation input during voice output. Since the prosody parameters are changed and the speech is synthesized and output based on the phoneme parameters and the prosody parameters, the user can partially emphasize the speech during the speech output, and can achieve high expressiveness. There is an effect that the present speech synthesizer can be used as a conversation alternative means having

According to the present invention, the prosody control parameter of the linguistic information is changed according to the operation input during the voice output, and the prosody parameter is generated from the linguistic information according to the prosody parameter generation rule. There is an effect that the prosody, such as a pitch pattern and a phoneme duration, is changed according to the operation of the user during the output of the voice, so that the user can partially emphasize the voice during the voice output.

According to the present invention, one prosody parameter generation rule is selected from a plurality of prosody parameter generation rules according to an operation input during voice output, and a prosody parameter is generated from linguistic information in accordance with the selected prosody parameter generation rule. With such a configuration, the prosody such as the pitch pattern and the phoneme duration is changed according to the user's operation during the output of the voice, and the user can partially emphasize the voice during the voice output. effective.

According to the present invention, the prosody parameters are generated from the linguistic information in accordance with the prosody parameter generation rules, and the generated prosody parameters are changed according to the first operation input during voice output. This has the effect that the user can instantly perform partial emphasis on the synthesized speech of.

According to the present invention, the boundary information having the timing of the accent phrase boundary and the syllable boundary is generated based on the prosody parameter and the linguistic information, and the changeable portion of the prosody parameter determined based on the boundary information is generated. , A portion whose prosody parameter is to be changed according to the second operation input during voice output is selected, and the prosody parameter of the portion is changed according to the first operation input during voice output. The prosody, such as the pitch pattern and the phoneme duration, is changed according to the user's operation during the output of, and the user can partially emphasize the voice during the voice output.

According to the present invention, a predetermined fixed message portion in the input text is detected, and voice waveform data of the fixed message of the type corresponding to the operation input is selected, and the voice waveform data corresponding to the operation input during the voice output is selected. The fixed message by changing its voice waveform data to change the prosody of the fixed message and play the fixed message as voice. This has the effect that the user can perform a general emphasis or the like.

According to the present invention, the input text is analyzed, and linguistic information including at least the accent phrase boundary position, accent type, reading and prosody control parameters corresponding to the input text is generated, and the linguistic information is input based on the linguistic information. A speech synthesis unit corresponding to the text is selected and used as a phonological parameter, a subsequent phonological parameter is changed according to an operation input during voice output, and a rhythm parameter is generated from linguistic information according to a rhythm parameter generation rule. And synthesized speech based on the prosody parameters, the user can partially emphasize the synthesized speech being output, etc., and use this speech synthesis as a conversational alternative with high expressiveness. There is an effect that the device can be used.

According to the present invention, one type of voice synthesis unit set is selected from a plurality of types of voice synthesis unit sets having different voice qualities and acoustic characteristics in accordance with an operation input during voice output, and the selected voice synthesis unit is selected. Since the speech synthesis unit corresponding to the input text is selected from the set based on the linguistic information and used as the phonological parameters, the voice quality and acoustic features such as gurgling and whispering are changed and the synthesized speech being output There is an effect that the user can immediately perform a partial emphasis on.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech synthesis device according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram illustrating a configuration example of a language processing unit in FIG. 1;

FIG. 3 is a diagram showing an example of a rhythm perception point table.

FIG. 4 is a diagram showing an example of a consonant duration table.

FIG. 5 is a diagram illustrating an example of an accent phrase string by a text analysis unit.

FIG. 6 is a diagram illustrating an example of language information by a language processing unit.

FIG. 7 is a diagram showing an example of a comparison of pitch patterns when a prosody control parameter (accent strength) is changed and when it is not changed.

FIG. 8 is a diagram showing an example of a comparison of phoneme continuation time lengths when there is a change in a prosody control parameter (average mora length) and when there is no change.

FIG. 9 is a block diagram showing a configuration of a speech synthesis device according to a second embodiment of the present invention.

FIG. 10 is a diagram showing an example of a plurality of types of conversion rules.

FIG. 11 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 3 of the present invention.

FIG. 12 is a diagram illustrating an example of a pitch pattern changed according to a prosody changing operation input.

FIG. 13 is a block diagram showing a configuration of a speech synthesis device according to a fourth embodiment of the present invention.

14 is a diagram showing boundary information /
It is a figure showing an example of boundary information generated by a prosody parameter generation part.

FIG. 15 is a diagram showing an example of a display screen of a display / input device for performing section input.

16 is a diagram showing an example of designation of a prosody change section in the display shown in FIG.

FIG. 17 shows a syllable string “everyone” when operated to increase the time length by 50% and when not operated.
FIG. 6 is a diagram showing an example of a phoneme duration time of the phoneme.

FIG. 18 is a block diagram showing a configuration of a speech synthesis device according to a fifth embodiment of the present invention.

FIG. 19 is a diagram illustrating an example of display by a speech synthesis unit set selection unit.

FIG. 20 is a block diagram showing a configuration of a speech synthesizer according to Embodiment 6 of the present invention.

21 is a block diagram illustrating a configuration example of a fixed message determination language processing unit in FIG. 20.

FIG. 22 is a diagram illustrating an example of fixed message data.

FIG. 23 is a diagram showing an example of voice waveform data numbers and comments displayed on the screen corresponding to the fixed message data of FIG. 22.

FIG. 24 is a diagram illustrating an example of a conversion rule when applying average power to a model.

FIG. 25 is a block diagram showing a configuration of a first conventional speech synthesizer.

FIG. 26 is a block diagram showing a configuration of a second conventional speech synthesizer.

[Explanation of symbols]

1 language processing unit, 3 phoneme parameter generation unit, 4 prosody control parameter change unit, 6 prosody parameter generation unit, 6
A boundary information / prosodic parameter generation unit (prosody parameter generation unit), 7 voice synthesis unit, 22 prosody parameter generation rule selection unit, 31 prosody parameter change unit, 31A
Section specifying prosody parameter changing section (prosodic parameter changing section), 42 voice synthesis unit set selecting section, 51 fixed message determination language processing section (language processing section), 53 fixed message selecting section, 54 fixed message prosody changing section, 5
5 voice reproduction unit, 81, 81A, 81B, 81C prosody parameter control unit, 82 phoneme parameter control unit.

Continued on the front page (72) Inventor Mitsuru Ebihara 2-3-2 Marunouchi, Chiyoda-ku, Tokyo Mitsubishi Electric Corporation F-term (reference) 5D045 AA09 AA20

Claims (11)

[Claims] A language processing unit that analyzes input text and generates linguistic information including at least an accent phrase boundary position, an accent type, a reading, and a prosody control parameter corresponding to the input text; A phoneme parameter generation unit for selecting a speech synthesis unit corresponding to the input text and using the phonetic parameter as a phoneme parameter; generating a prosody parameter from the linguistic information according to a prosody parameter generation rule; A speech synthesis device comprising: a prosody parameter control unit that changes a parameter; and a speech synthesis unit that synthesizes and outputs speech based on the phoneme parameter and the prosody parameter. 2. A prosody parameter control unit for changing a prosody control parameter of the linguistic information according to an operation input during voice output, and a prosody parameter changing unit based on the prosody parameter generation rule. The speech synthesis apparatus according to claim 1, further comprising a prosody parameter generation unit that generates the prosody. 3. A prosody parameter generation rule selection unit that selects one prosody parameter generation rule from a plurality of prosody parameter generation rules in accordance with an operation input during voice output, and a prosody parameter generation rule selection unit. A prosody parameter generation unit configured to generate a prosody parameter from linguistic information in accordance with the prosody parameter generation rule selected by the unit.
A speech synthesizer as described. 4. A prosody parameter control unit configured to generate a prosody parameter from linguistic information in accordance with a prosody parameter generation rule, and a first prosody parameter generated by the prosody parameter generation unit that outputs the prosody parameter in a speech output. 2. The speech synthesizer according to claim 1, further comprising a prosody parameter changing unit that changes in accordance with an operation input. 5. A prosody parameter generation unit generates boundary information having timing of an accent phrase boundary and a syllable boundary based on the generated prosody parameter and linguistic information, and a prosody parameter change unit based on the boundary information. A part for changing the prosody parameter according to the second operation input during voice output is selected from the determined prosody parameter changeable part, and the prosody parameter of the part is selected according to the first operation input during voice output. 5. The voice synthesizing device according to claim 4, wherein the voice synthesizing unit changes the frequency. 6. A language processing unit detects a predetermined fixed message part of the input text, analyzes a part other than the fixed message, and detects an accent phrase boundary position corresponding to the part other than the fixed message. A fixed message selection unit that generates linguistic information including at least accent type, reading and prosody control parameters, and selects voice waveform data of the detected fixed message in accordance with an operation input during voice output, and an operation input during voice output 2. The apparatus according to claim 1, further comprising: a fixed message prosody change unit configured to change the prosody of the fixed message by changing the voice waveform data in accordance with a voice message, and an audio reproduction unit configured to reproduce the fixed message as voice. The speech synthesizer according to claim 4. 7. Analyzing the input text and generating linguistic information including at least an accent phrase boundary position, an accent type, a reading and a prosody control parameter corresponding to the input text, and the input text based on the linguistic information. Selecting a speech synthesis unit corresponding to a rhythm parameter and generating a rhythm parameter from the linguistic information according to a rhythm parameter generation rule, and changing a subsequent rhythm parameter in accordance with an operation input during voice output. And a step of synthesizing and outputting speech based on the phoneme parameter and the prosodic parameter. 8. A language processing unit that analyzes input text and generates linguistic information including at least an accent phrase boundary position, an accent type, a reading, and a prosody control parameter corresponding to the input text. While selecting a speech synthesis unit corresponding to the input text and setting it as a phoneme parameter,
A phoneme parameter control unit that changes a subsequent phoneme parameter according to an operation input during voice output, a prosody parameter generation unit that generates a prosody parameter from the linguistic information according to a prosody parameter generation rule, And a speech synthesizer that synthesizes and outputs speech based on the speech. 9. A speech synthesis unit set for selecting one type of speech synthesis unit set according to an operation input during speech output from a plurality of types of speech synthesis unit sets having different voice qualities and acoustic characteristics. A selecting unit, and a phoneme parameter generation unit that selects a speech synthesis unit corresponding to the input text based on linguistic information from the speech synthesis unit set selected by the speech synthesis unit set selection unit, and sets the selected speech synthesis unit as a phoneme parameter. 9. The speech synthesizer according to claim 8, wherein: 10. A language processing unit detects a predetermined fixed message part of the input text, analyzes a part other than the fixed message, and detects an accent phrase boundary position corresponding to the part other than the fixed message. A fixed message selection unit that generates linguistic information including at least accent type, reading and prosody control parameters, and selects voice waveform data of the detected fixed message in accordance with an operation input during voice output, and an operation input during voice output 9. A fixed message prosody change unit that changes the prosody of the fixed message by changing the audio waveform data according to the following, and a voice reproduction unit that reproduces the fixed message as voice. The speech synthesizer according to claim 9. 11. An input text is analyzed, and an accent phrase boundary position corresponding to the input text, an accent type,
Generating linguistic information including at least reading and prosodic control parameters, and selecting a speech synthesis unit corresponding to the input text based on the linguistic information as a phonological parameter,
Changing a subsequent phonological parameter according to an operation input during voice output; generating a prosodic parameter from the linguistic information according to a prosodic parameter generation rule; synthesizing a voice based on the phonological parameter and the prosodic parameter And outputting.