JPH06250685A - Voice synthesis system and rule synthesis device - Google Patents

Abstract

PURPOSE:To obtain a voice synthesis system and a rule synthesis device which are applicable to a rule synthesis that synthesizes high quality voice from character strings based on rules. CONSTITUTION:The system consists of a voice sound source generating means 10 which generates a voice sound source 104 that repeats an inputted one pitch length residual waveform 107 at a frame average pitch 100 interval, a no voice sound source generating means 11, a fluctuation component generating means 13 which generates fluctuation components 200 that are normalized by a frame average power 101, a fluctuation component superimposing means 14 which outputs a fluctuation component superimposed voice sound source 201 by superimposing the waveform 104 equivalent to one pitch to the fluctuation components 200 and a vocal tract filter means 12 which outputs synthesized voices 206 by the filter that simulates vocal tract characteristics 103 of LSP filters by a no voice sound source 105 from the source 201 or the means 11.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing method and a rule synthesizing apparatus, and more particularly to a speech synthesizing method and a rule synthesizing apparatus applied to rule synthesizing for converting a sentence given by characters into speech.

[0002]

2. Description of the Related Art Speech synthesis (speech synthesis) is a method of artificially producing speech without directly relying on human speech.
Synthesis). The voice synthesis method can be classified into three types, that is, a recording editing method, a parameter editing method, and a rule synthesizing method. Among them, the recording editing method and the parameter editing method are methods in which the voice of a person stored in advance is directly connected and output, so that there is an advantage that a voice close to natural can be synthesized. There is a problem that the vocabulary and sentence structure that can be output are limited. However, even if the type, sentence structure, vocabulary, etc. of the response sentence to be output vary greatly depending on the user, it cannot be said to be true voice synthesis unless the quality of synthesized speech is deteriorated. In this respect, the rule synthesis method is a method for producing speech from a character string or a phoneme symbol string based on phonetic or linguistic rules. Since the number of pronunciation vocabularies can be infinite, the rule synthesis method is essentially It can be said that there is a possibility that true voice synthesis can be performed, unlike the recording editing method and the parameter editing method.

The principle of speech synthesis by such a characteristic rule synthesis method uses a basic small unit of special tone parameter such as a syllable, a phoneme, a waveform of one pitch interval, etc. as a unit to be stored, and instead thereof. By precisely defining rules for connecting them and rules for controlling temperament information such as pitch and amplitude, any word can be synthesized from a phoneme, a syllable symbol, or a series of characters. At this time, in order for the voice to be natural and easy to hear, changes in pitch and stress, and temporal changes in spectrum must be smooth, and poses must be natural. Therefore, in the case of this rule synthesis method, the control rule (control information and control mechanism) of the acoustic parameters, which is based on the phonetic or linguistic characteristics of natural speech, plays an important role in addition to the quality of the basic unit used for synthesis. Fulfill.

By the way, a vocoder method has been conventionally known as a speech synthesizing method which can be applied to such rule synthesizing and obtains a synthetic speech having an arbitrary pitch period, power and duration. The vocoder method is, for example, as shown in "Digital Speech Processing" Sadahiro Furui Tokai University Press, based on the analysis result of the speech signal, the speech is separated into sound source information and vocal tract information to be modeled as synthetic speech. This is a method of obtaining the synthesized speech with a desired pitch period, which is relatively easy.

FIG. 8 is a block diagram showing an example of the configuration of a conventional vocoder-based voice analysis / synthesis system for performing voice analysis / synthesis using linear prediction analysis. In the figure, a voice analysis / synthesis system based on a conventional vocoder system is shown in FIG.
A frame average pitch 100, which is sound source information obtained by separating and modeling the input voice according to the analysis result of the voice signal.
, Frame average power 101, voiced unvoiced information 102
And a vocal tract characteristic 103, which is vocal tract information obtained by separating and modeling the input voice, are input, and a voiced sound source 104
And the unvoiced sound source 105 are intermediately output, and finally the synthetic speech 106 is output.

The operation of the conventional voice analysis / synthesis system based on the vocoder system configured as described above will be described. The voiced sound source generation means 10 generates a voiced sound source 104 represented by an impulse train having a constant frame average pitch interval by the frame average power 101 and the frame average pitch 100 in the voiced section determined by the voiced unvoiced information 102. Further, the unvoiced sound source generation means 11 generates the unvoiced sound source 105 represented by white noise with the frame average power 101 in the unvoiced section determined by the voiced unvoiced information 102. The vocal tract filter means 12 drives a vocal tract filter that approximates the vocal tract characteristic 103 with the voiced sound source 104 or the unvoiced sound source 105, and outputs a synthetic speech 106.

[0007]

In the conventional vocoder system as described above, since the impulse train is used as the sound source, there is a problem in that the minute temporal characteristics of the pitch intervals of the voiced sound are lost. When the estimation of vocal tract characteristics is insufficient, the fine feature of the spectrum is lost, so that the quality of synthesized speech is deteriorated. In order to solve this problem, as a method improved so that minute features of the voice can be left, a method of using a residual waveform obtained by analyzing the voice with an inverse filter as a sound source or approximating the residual waveform A multi-pulse method using a sound source has been proposed.

However, although it has become possible to obtain a sufficiently high quality synthesized voice as compared with these improved vocoder systems, on the other hand, there is a problem that the amount of data to be stored becomes enormous, and further, that of voice Fine time characteristics tend to depend on power and pitch.In these methods, when applying rule synthesis, the fine time characteristics of the original speech are preserved as they are in the synthesized speech. It has been pointed out that there is a problem with it.

On the other hand, as a method not using a filter,
A waveform superposition method that enables control of the pitch period by direct expression of the speech waveform has been proposed, and it has become possible to realize high-quality synthesized speech. However, this method also has the same problem as the method using the residual waveform as a sound source in the application to rule synthesis.

Further, a method of superposing a fluctuation component of a natural voice or a natural sound source stored in advance on a synthetic sound or a synthetic sound source or a method of superposing a fluctuation component generated by a random number on a synthetic sound source has been proposed. Similar to the above methods, neither method considers that the fluctuation component tends to depend on power or pitch,
There is a problem in that an appropriate fluctuation component is not generated for a desired power or pitch.

The present invention has the above-mentioned problems, that is, the amount of data to be stored in the method for directly storing the minute features of the voice becomes huge, and the minute time features are power and pitch. It was made to solve the problem that the same characteristics as the original are not exhibited when changing, and it is to provide a high-quality speech synthesis method that can be applied to rule synthesis. Has an aim.

[0012]

In order to achieve the above object, a speech synthesis method according to the first aspect of the present invention uses a vocal tract characteristic, a frame average power, a frame average pitch, a voiced voice obtained by analyzing a voice. An unvoiced sound source generation unit that inputs unvoiced information and generates an unvoiced sound source represented by white noise,
A voice synthesis method having a voiced sound source generation means for generating a voiced sound source, which is a repetition of a 1-pitch length residual waveform representing an impulse or a frame for each frame average pitch, wherein a frame average power or a frame average pitch is input, A fluctuation component generating means for generating a fluctuation component defined by power or pitch; and a fluctuation component superposing means for superposing the fluctuation component obtained by the fluctuation generating means on a voiced sound source waveform for one pitch for each frame average pitch section, Vocal tract filter means for obtaining synthetic speech by a filter approximating the vocal tract characteristics, using the fluctuation component superimposed sound source waveform obtained by the fluctuation component superimposing means as an input.

The rule synthesizing device according to the second invention is
A rule synthesizing device that inputs a character string or a phoneme symbol string and outputs synthetic speech according to phonetic or linguistic rules based on pre-stored dictionary information or speech unit information. A fluctuation component generating means for generating a fluctuation component defined by power or pitch generated based on a rule, and a fluctuation component obtained by superposing the fluctuation component obtained by the fluctuation generating means on a voiced sound source waveform for one pitch for each pitch section. The component superimposing means and the fluctuation component superimposing sound source waveform obtained by the fluctuation component superimposing means are input,
Vocal tract filter means for obtaining synthesized speech with a filter approximating the vocal tract characteristic generated by the rule.

The voice synthesis system according to the third invention is
A frame average power or a frame average pitch is input for a voiced sound section that is determined by voiced unvoiced information, and a fluctuation component generation unit that generates a fluctuation component defined by the power or pitch, and a fluctuation component obtained by the fluctuation generation unit, Fluctuation component superimposing means that superimposes on the vocal tract characteristics of each frame in synchronization with the frame average pitch, and voiced sound source waveforms are input for each pitch section, and the fluctuation component superimposing vocal tract characteristics obtained by the fluctuation component superimposing means Vocal tract filter means for obtaining synthetic speech by an approximate filter.

Further, the rule synthesizing device according to the fourth invention is
A fluctuation component generating means for generating a fluctuation component defined by power or pitch generated based on a rule according to a sentence, and a fluctuation component obtained by the fluctuation generating means,
Fluctuation component superimposing means that superimposes on the vocal tract characteristics of each frame in the voiced section generated by the rule in synchronization with the frame average pitch, and a voiced sound source waveform is input for each pitch section and obtained by the fluctuation component superimposing means. And a vocal tract filter means for obtaining a synthetic voice by a filter approximating the vocal tract characteristic with the superimposed fluctuation component.

The speech synthesis system according to the fifth aspect of the invention is
A fluctuation component generating means for inputting the frame average power or the frame average pitch to generate a fluctuation component defined by the power or the pitch, and a fluctuation component obtained by the fluctuation generating means for one pitch long speech waveform for each frame average pitch section. And a waveform superimposing means for obtaining a synthesized speech by superimposing the fluctuation component superimposing speech waveform of one pitch length obtained by the fluctuation component superimposing means at a frame average pitch interval. Characterize.

Further, the rule synthesizing apparatus according to the sixth aspect of the invention is obtained from the fluctuation component generating means for generating the fluctuation component defined by the power or pitch generated based on the rule according to the sentence, and the fluctuation generating means. A fluctuation component superimposing means for superimposing the fluctuation component on a pitch-based 1-pitch long speech waveform for each pitch section, and a fluctuation component superimposing speech waveform of 1 pitch length obtained by the fluctuation component superimposing means are superimposed at pitch intervals. And a waveform superimposing means for obtaining a synthetic voice by doing so.

[0018]

Therefore, according to the voice synthesizing method of the first aspect of the present invention, the fluctuation component generating means generates the fluctuation component defined by the power or pitch of the input voice, and the fluctuation component superposing means generates the fluctuation component. A fluctuation component is superimposed on a voiced sound source waveform for one pitch for each frame average pitch section, and the fluctuation component superimposed sound source waveform is input so that the vocal tract filter means obtains a synthetic speech by a filter that approximates the vocal tract characteristics. Therefore, a synthetic voice with high naturalness is reproduced.

According to the rule synthesizing device of the second aspect of the invention, the fluctuation component generating means generates the fluctuation component defined by the power or the pitch, and the fluctuation component superposing means generates the fluctuation component by one pitch for each pitch section. Since the vocal tract filter means for inputting the fluctuation component-superposed source waveform is superposed on the voiced voicing source waveform for a minute, the synthesized voice is uttered by a filter that approximates the vocal tract characteristics generated by the rule, so that high quality is achieved. The rule synthesized sound of is obtained.

Further, according to the voice synthesizing method of the third aspect of the invention, the fluctuation component generating means generates a fluctuation component defined by the power or pitch of the input voice, and the fluctuation component superposing means converts the fluctuation component. Superimpose on the vocal tract characteristics of each frame in synchronization with the frame average pitch,
Since the vocal tract filter means for inputting the voiced sound source waveform for each pitch section obtains the synthetic voice by the filter that approximates the fluctuation component superimposed vocal tract characteristic, the synthetic voice with high naturalness is reproduced.

Further, according to the rule synthesizing apparatus of the fourth aspect of the invention, the fluctuation component generating means generates the fluctuation component defined by the power or the pitch, and the fluctuation component superposing means generates the fluctuation component by the voiced voice. Vocal tract filter means for superimposing on the vocal tract characteristic of each frame in the section in synchronization with the frame average pitch and inputting a voiced sound source waveform for each pitch section is synthesized by a filter approximating the fluctuation component superimposed vocal tract characteristic. As a result, a high-quality regular synthesized voice can be obtained.

According to the voice synthesizing method of the fifth aspect of the invention, the fluctuation component generating means generates a fluctuation component defined by the power or pitch of the input voice, and the fluctuation component superposing means converts the fluctuation component. Since the 1-pitch-long voice waveform is superimposed on each frame average pitch section and the fluctuation component-superimposed voice waveform of the 1-pitch length is superimposed by the waveform superimposing means at the frame average pitch interval, synthetic speech is obtained. A synthetic voice with high naturalness is reproduced.

Further, according to the rule synthesizing device of the sixth aspect of the present invention, the fluctuation component generating means generates the fluctuation component defined by the power or the pitch, and the fluctuation component superposing means regulates the fluctuation component for each pitch section. Since the synthesized voice is generated by superposing it on the generated one-pitch-long voice waveform and superposing the fluctuation component-superposed voice waveform of the one-pitch length at pitch intervals, a high-quality rule is produced. A synthetic sound is obtained.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A preferred embodiment of the present invention will be described below with reference to the drawings. 1 to 6 are configuration diagrams showing a configuration example of a voice analysis / synthesis system by a voice synthesis system according to the present embodiment or a rule synthesis device according to the present embodiment. In the figure,
The same or corresponding parts as those of the conventional voice analysis / synthesis system by the voice synthesis method are designated by the same reference numerals, and the description thereof will be omitted.

Example 1. FIG. 1 is a configuration diagram showing a configuration example of a first voice analysis / synthesis system by a voice synthesis system according to the present embodiment. In the figure, the first speech analysis and synthesis system of the present embodiment includes a voiced sound source generation means 10, an unvoiced sound source generation means 11, a vocal tract filter means 12, and a fluctuation component generation means 13
And fluctuation component superimposing means 14. Further, a frame average pitch 100, a frame average power 101, voiced unvoiced information 102, which is sound source information obtained by separating and modeling the input voice based on the analysis result of the voice signal, and a vocal tract obtained by separating and modeling the input voice in the same manner. In addition to the vocal tract characteristic 103, which is information, the 1-pitch length residual waveform 107 is input, and in addition to the voiced sound source 104 and the unvoiced sound source 105, the fluctuation component 200 and the fluctuation component-superimposed voiced sound source 201 are intermediately output, and Finally, the synthesized voice 206 is output.

Next, the operation of the first speech analysis and synthesis system of this embodiment having the above-mentioned configuration will be described. The fluctuation component generation means 13 inputs the frame average power 101 and outputs a fluctuation component 200 defined by the power as shown in FIG. 7, for example. 7. FIG. 7 is a graph showing the relationship between the frame average power and the fluctuation component.

The unvoiced sound source generating means 11 uses the voiced unvoiced information 1
When it is determined to be unvoiced by 02, the unvoiced sound source 105 including white noise is generated. The voiced sound source generation means 10 generates a voiced sound source 104 by repeating the 1-pitch length residual waveform 107 at frame average pitch 100 intervals when it is determined to be voiced by the voiced unvoiced information 102. The fluctuation component superimposing means 14 superimposes the fluctuation component 200 on the voiced sound source waveform 104 for one pitch, and outputs the fluctuation component superimposed voiced sound source 201. The vocal tract filter means 12 receives the voiced sound source 201 or the unvoiced sound source 105 on which the fluctuation component is superimposed, and obtains a synthetic speech 206 by a filter that approximates the vocal tract characteristic 103 such as an LSP filter.

Example 2. FIG. 2 is a configuration diagram showing a configuration example of the first rule synthesis device according to the present embodiment. In the figure, the first rule synthesizing apparatus 20 of the present embodiment is composed of a sentence analyzing unit 21, a synthesizing rule unit 22, and a voice synthesizing unit 23, and the voice synthesizing unit 23 is a voiced sound source generating unit 10.
Unvoiced sound source generation means 11 and vocal tract filter means 12
, Fluctuation component generation means 13 and fluctuation component superposition means 1
4 and. Then, the first rule synthesizing device 20 uses the input sentence 2 composed of a character string or a phoneme symbol string.
02 is input, and dictionary information 203 and speech unit information 204
Rule composition is performed using and.

Next, the operation of the first rule synthesizing device of the present embodiment configured as described above will be described. Input sentence 2
In the rule synthesizing device 20 that inputs 02 and outputs the synthesized speech 206, the sentence analysis unit 21 stores the input sentence 202 given in characters in the dictionary information 20 stored in advance.
3 is referred to and analyzed, and a sentence analysis result 205 such as word reading and accent is output. The synthesis rule means 22 inputs the sentence analysis result 205 from the sentence analysis means 21, refers to the pre-stored speech segment information 204, and according to the rule, the 1-pitch length residual waveform 107 / voice used for speech synthesis. Road characteristic 103, frame average power 101,
The frame average pitch 100 and voiced / unvoiced information 102 are determined and output.

In the voice synthesizing means 23, the 1-pitch length residual waveform 107, the vocal tract characteristic 103, the frame average power 101, the frame average pitch 100 and the voiced unvoiced information 102 are input, and the voice synthesizing means 23 generates an unvoiced sound source. The means 11 generates an unvoiced sound source 105 composed of white noise when it is determined to be unvoiced by the voiced unvoiced information 102. Further, the voiced sound source generation means 10 repeats the 1-pitch length residual waveform 107 at frame average pitch 100 intervals when it is determined that the voiced unvoiced information 102 is voiced.
04 is generated. The fluctuation component generation means 13 inputs the frame average power 101 and outputs a fluctuation component 200 defined by the power as shown in FIG. 7, for example.
The fluctuation component superimposing means 14 superimposes the fluctuation component 200 on the voiced sound source 104 for one pitch for every 100 frame average pitches, and outputs the fluctuation component superimposed voiced sound source 201. The vocal tract filter means 12 receives the voiced sound source 201 or the unvoiced sound source 105 on which the fluctuation component is superimposed, and obtains a synthetic speech 206 by a filter such as an LSP filter that approximates the vocal tract characteristic 103.

Example 3. FIG. 3 is a configuration diagram showing a configuration example of the second voice analysis and synthesis system by the voice synthesis method according to the present embodiment. In the figure, the second speech analysis and synthesis system of the present embodiment includes a voiced sound source generation means 10, an unvoiced sound source generation means 11,
It has a vocal tract filter means 12, a fluctuation component generation means 13, and a fluctuation component superposition means 14. Further, frame average pitch 100, frame average power 101, voiced unvoiced information 102, vocal tract characteristics 103, 1 pitch length residual waveform 107
Is input, the voiced sound source 104, the unvoiced sound source 105, the fluctuation component 200, and the fluctuation component superimposed vocal tract characteristic 207 are intermediately output, and finally the synthesized speech 206 is output. Such a configuration is similar to the configuration of the first voice analysis / synthesis system of the present embodiment shown in FIG. 1, but the fluctuation component superimposing means 14 of the second voice analysis / synthesis system shifts the fluctuation component 200 to the frame average pitch. A vocal tract characteristic 207 is output by superimposing the fluctuation component in synchronism with 100, and the vocal tract filter means 12 outputs this.
It is characterized in that the synthetic speech 206 is generated by inputting into the input.

Next, the operation of the second speech analysis and synthesis system of the present embodiment configured as described above will be described. The fluctuation component generation means 13 inputs the frame average power 101 and outputs a fluctuation component 200 defined by the power as shown in FIG. 7, for example. The unvoiced sound source generator 11
An unvoiced sound source 105 made of white noise is generated when it is determined to be unvoiced by the voiced unvoiced information 102. The voiced sound source generation means 10 generates a voiced sound source 104 by repeating the 1-pitch length residual waveform 107 at frame average pitch 100 intervals when it is determined to be voiced by the voiced unvoiced information 102.
Further, in the voiced state, the fluctuation component superimposing unit 14 superimposes the fluctuation component 200 on the impulse response of the vocal tract characteristic 103 in synchronization with the frame average pitch 100, and outputs the fluctuation component superimposed vocal tract characteristic 207. Vocal tract filter means 12
Inputs a voiced sound source 104 or an unvoiced sound source 105, and obtains a synthesized voice 206 by a filter approximating a fluctuation component superimposed vocal tract characteristic 207.

Example 4. FIG. 4 is a configuration diagram showing a configuration example of the second rule synthesis device according to the present embodiment. In the figure, the second rule synthesizing device 20 of the present embodiment is composed of a sentence analyzing means 21, a synthesizing rule means 22, and a voice synthesizing means 23, and the voice synthesizing means 23 is a voiced sound source generating means 10.
Unvoiced sound source generation means 11 and vocal tract filter means 12
, Fluctuation component generation means 13 and fluctuation component superposition means 1
4 and. Then, the second rule synthesis device 20 inputs the input sentence 202, and inputs the dictionary information 203,
Rule synthesis is performed using the speech unit information 204.

Next, the operation of the second rule synthesizing device of this embodiment constructed as described above will be explained. Input sentence 2
In the rule synthesizing device 17 that inputs 02 and outputs the synthesized speech 106, the sentence analysis unit 21 converts the input sentence 202 given in characters into the previously stored dictionary information 20.
3 is referred to and analyzed, and a sentence analysis result 205 such as word reading and accent is output. The synthesis rule means 22 inputs the sentence analysis result 205, refers to the voice segment information 204 stored in advance, and uses the rule to generate a 1-pitch length residual waveform 107, vocal tract characteristic 103, and frame average used for voice synthesis. Power 101 · Frame average pitch 100 ·
The voiced / unvoiced information 102 is determined and output to the voice synthesis unit 23.

In the voice synthesizing means 23, the 1-pitch length residual waveform 107, the vocal tract characteristic 103, the frame average power 101, the frame average pitch 100, and the voiced unvoiced information 102 are input, and the voice synthesizer 23 generates an unvoiced sound source. The means 11 generates an unvoiced sound source 105 composed of white noise when it is determined to be unvoiced by the voiced unvoiced information 102. Further, the voiced sound source generation means 10 repeats the 1-pitch length residual waveform 107 at frame average pitch 100 intervals when it is determined that the voiced unvoiced information 102 is voiced.
04 is generated. The fluctuation component generation means 13 inputs the frame average power 101 and outputs the fluctuation component 200 defined by the power as shown in FIG. 7, for example, and the fluctuation component superposition means 14 averages the fluctuation component 200 by the frame average. It is superimposed on the impulse response of the vocal tract characteristic 103 in synchronization with the pitch 100, and the fluctuation component superimposed vocal tract characteristic 207 is output. The vocal tract filter means 12 inputs the voiced sound source 104 or the unvoiced sound source 105, and the fluctuation component-superimposed vocal tract characteristic 2
A synthesized voice 206 is obtained by a filter approximating 07.

Example 5. FIG. 5 is a configuration diagram showing a configuration example of a third voice analysis / synthesis system by the voice synthesis method according to the present embodiment. In the figure, the third voice analysis / synthesis system of the present exemplary embodiment includes a fluctuation component generation means 13 and a fluctuation component superposition means 1.
4. It has a waveform superimposing means 15. Further, the frame average pitch 100, the frame average power 101, and the 1-pitch long speech waveform 208 are input, the fluctuation component 200 and the fluctuation component superimposed speech waveform 209 are intermediately output, and finally the synthesized speech 206 is output.

Next, the operation of the third speech analysis / synthesis system of the present embodiment having the above configuration will be described. The fluctuation component generation means 13 inputs the frame average power 101 and outputs a fluctuation component 200 defined by the power as shown in FIG. 7, for example. Fluctuation component superimposing means 14
When the voiced unvoiced information determines that the voice is voiced, the fluctuation component 200 from the fluctuation component generation means 13 is superimposed on the 1-pitch long speech waveform 208, and the fluctuation component superimposed speech waveform 209 is output. The waveform superimposing unit 15 receives the fluctuation component-superimposed speech waveform 209 and superimposes it at a frame average pitch of 100 to obtain a synthesized speech 206.

Example 6. FIG. 6 is a configuration diagram showing a configuration example of the third rule synthesizing device according to the present embodiment. In the figure, the third rule synthesizing device 20 of the present embodiment is composed of a sentence analyzing means 21, a synthesizing rule means 22, and a voice synthesizing means 23, and the voice synthesizing means 23 is a fluctuation component generating means 13 and Fluctuation component superimposing means 14 and waveform superimposing means 15
It consists of and. Then, the third rule synthesizer 2
For 0, the input sentence 202 is input, and rule synthesis is performed using the dictionary information 203 and the voice segment information 204.

Next, the operation of the third rule synthesizing device of the present embodiment configured as described above will be described. Input sentence 2
In the rule synthesizing device 20 that inputs 02 and outputs the synthesized speech 206, the sentence analysis unit 21 stores the input sentence 202 given in characters in the dictionary information 20 stored in advance.
3 is referred to and analyzed, and a sentence analysis result 205 such as word reading and accent is output. The synthesis rule means 22 inputs the sentence analysis result 205 from the sentence analysis means 21, refers to the pre-stored speech segment information 204, and according to the rule, a 1-pitch long speech waveform 208 / frame average used for speech synthesis. The pitch 100 and the frame average power 101 are determined and output to the voice synthesizing means 23.

In the voice synthesizing means 23, the 1-pitch long voice waveform 208, the frame average pitch 100, and the frame average power 101 are input, and the fluctuation component generating means 13 of the voice synthesizing means 23 inputs the frame average power 101. For example, the fluctuation component 200 defined by the power as shown in FIG. 7 is output. Further, the fluctuation component superimposing means 14 superimposes the fluctuation component 200 on the 1-pitch long speech waveform 208 and outputs a fluctuation component-superimposed speech waveform 209. The waveform superimposing means 15 inputs the fluctuation component-superimposed speech waveform 209 and superimposes it at intervals of the frame average pitch 100 to obtain a synthesized speech 206.

Other Embodiments In the description of the fluctuation component generation means 13 in the above-described first to sixth embodiments, the fluctuation component generation means 13 inputs the frame average power 101 and outputs the fluctuation component 200 defined by the power. However, the present invention is not limited to this. For example, the frame average pitch may be input and a fluctuation component defined by the pitch may be output.

Further, in the description of the voiced sound source generation means 10 of the above-described first to fourth embodiments, the voiced sound source generation means 10 is described.
Is a 1-pitch length residual waveform 107 when it is determined to be voiced by the voiced unvoiced information 102.
Although it has been described that the voiced sound source 104 that is repeated at intervals is generated, the present invention is not limited to this, and for example, an impulse train generated by the input power and pitch can be used.

[0043]

As described above, according to the voice synthesizing method of the present invention, the fluctuation component defined by the power or pitch of the input voice is generated, and the fluctuation component is used for the pitch length. The voiced sound source waveform or the vocal tract characteristics of the voiced section is superposed at each pitch interval, or the voiced speech waveform superposed at each pitch interval is configured to be superposed at each pitch interval. It has the effect that it can be realized.

Further, according to the rule synthesizing apparatus of the present invention, the fluctuation component defined by the power or the pitch generated by the rule is generated, and the fluctuation component is used for the synthesis of the pitched voiced sound source waveform or voiced sound source. Since it is configured to superimpose on the vocal tract characteristics of a section at each pitch interval, or to superimpose a waveform obtained by superimposing on a pitch-length voiced speech waveform at each pitch interval, a high quality rule. There is an effect that a synthetic sound can be obtained.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a configuration example of a first voice analysis / synthesis system by a voice synthesis method according to an embodiment.

FIG. 2 is a configuration diagram showing a configuration example of a first rule synthesis device according to the present embodiment.

FIG. 3 is a configuration diagram showing a configuration example of a second voice analysis / synthesis system by a voice synthesis system according to the present embodiment.

FIG. 4 is a configuration diagram showing a configuration example of a second rule synthesizing device according to the present embodiment.

FIG. 5 is a configuration diagram showing a configuration example of a third voice analysis / synthesis system by the voice synthesis system according to the present embodiment.

FIG. 6 is a configuration diagram showing a configuration example of a third rule composition device according to the present embodiment.

FIG. 7 is a graph showing the relationship between the frame average power and the fluctuation component generated by the fluctuation component generation means.

FIG. 8 is a configuration diagram showing a configuration example of a voice analysis / synthesis system according to a conventional voice synthesis system.

[Description of Codes] 10 voiced sound source generation means 11 unvoiced sound source generation means 12 vocal tract filter means 13 fluctuation component generation means 14 fluctuation component superposition means 15 waveform superposition means 20 rule synthesis device 21 sentence analysis means 22 synthesis rule means 23 speech synthesis means 100 frame average pitch 101 frame average power 102 voiced unvoiced information 103 vocal tract characteristic 104 voiced sound source 105 unvoiced sound source 106, 206 synthesized speech 107 1 pitch length residual waveform 200 fluctuation component 201 voiced sound source 202 input sentence 203 dictionary information 204 Speech segment information 205 Sentence analysis result 207 Fluctuation component superimposed vocal tract characteristics 208 1 pitch length speech waveform 209 Fluctuation component superimposed speech waveform

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 19, 1993

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] 0002

[Correction method] Change

[Correction content]

[0002]

2. Description of the Related Art Speech synthesis (speech synthesis) is a method of artificially producing speech without directly relying on human speech.
Synthesis). The voice synthesizing method can be classified into three types, that is, a recording editing method, a parameter editing method, and a rule synthesizing method. Among them, the recording editing method and the parameter editing method are methods in which the voice of a person stored in advance is directly connected and output, so that there is an advantage that a voice close to natural can be synthesized. There is a problem that the vocabulary and sentence structure that can be output are limited . Point This <br/>, the rule synthesizing method is a method of creating a sound based on a character string or phoneme symbol string to phonetically to linguistic rules, Recording editing method and parameters editing method and a different < It is possible to synthesize speech of any vocabulary with a small storage capacity.
It

[Procedure Amendment 2]

[Document name to be amended] Statement

[Name of item to be corrected] 0003

[Correction method] Change

[Correction content]

[0003] The rule synthesizing method shown here, the syllable as stored in advance unit, a phoneme, such as a waveform of one pitch period, using the feature parameters of the basic small unit, rules for connecting Instead them and, by determining precisely the rules governing the rhyme <br/> law information, such as pitch and amplitude, in any words, phonemes, it is intended to be able to synthesize the syllabic symbol or character sequence. At this time,
For the voice to be natural and easy to hear, changes in pitch and stress, and temporal changes in the spectrum must be smooth, and poses must be natural. Therefore, in the case of the rule synthesizing method, basic for use in the synthesis
With quality of the single position, based on the phonetic or linguistic characteristics of natural speech, control rules of acoustic parameters (control information and control mechanism) play an important role.

Claims (6)

[Claims] 1. An unvoiced sound source generation means for generating an unvoiced sound source represented by white noise by inputting vocal tract characteristics, frame average power, frame average pitch, and voiced unvoiced information obtained by speech analysis, and impulse or frame. In a voice synthesis method having a voiced sound source generation means for generating a voiced sound source consisting of repetitions of a 1-pitch length residual waveform for each frame average pitch, the frame average power or the frame average pitch is input and defined by the power or pitch. And a fluctuation component superimposing means for superimposing the fluctuation component obtained by the fluctuation generating means on the voiced sound source waveform for one pitch for each frame average pitch section, and the fluctuation component superimposing means. Using the obtained fluctuation source superimposed source waveform as an input, a filter that approximates the vocal tract characteristics Speech synthesis method characterized by comprising the vocal tract filter means for obtaining a synthesized speech, the by motor. 2. A rule synthesizing device for inputting a character string or a phoneme symbol string, and outputting synthetic speech according to phonetic or linguistic rules based on prestored dictionary information, phoneme segment information, or the like, A fluctuation component generating means for generating a fluctuation component defined by power or pitch generated based on a rule according to a sentence, and a fluctuation component obtained by the fluctuation generating means are converted into a voiced sound source waveform for one pitch for each pitch section. A fluctuation component superimposing means for superimposing, and a vocal tract filter means for receiving a fluctuation component superimposing sound source waveform obtained by the fluctuation component superimposing means and obtaining a synthesized voice by a filter approximating the vocal tract characteristics generated by the rule. A rule synthesizer characterized in that 3. An unvoiced sound source generation means for generating an unvoiced sound source represented by white noise, using vocal tract characteristics, frame average power, frame average pitch, and voiced unvoiced information obtained by speech analysis as inputs, and impulse or frame. In a voice synthesis method having a voiced sound source generating means for generating a voiced sound source consisting of repetitions of a 1-pitch length residual waveform for each frame average pitch, a frame average power or frame for a voiced sound section determined by voiced unvoiced information. A fluctuation component generation means for inputting an average pitch and generating a fluctuation component defined by power or pitch, and a fluctuation component obtained by the fluctuation generation means are superimposed on the vocal tract characteristics of each frame in synchronization with the frame average pitch. Fluctuation component superimposing means and input of voiced source waveform for each pitch section Speech synthesis method characterized by comprising, a vocal tract filter means for obtaining a synthesized speech by filtering which approximates the fluctuation component superimposed vocal tract characteristics obtained by the fluctuation component superposing unit. 4. A rule synthesizing device for inputting a character string or a phoneme symbol string and outputting synthetic speech in accordance with phonetic or linguistic rules based on prestored dictionary information, phoneme segment information, or the like. A fluctuation component generating means for generating a fluctuation component defined by power or pitch generated based on a rule according to a sentence, and a fluctuation component obtained by the fluctuation generating means are used as a voice of each frame of a voiced section generated by the rule. Fluctuation component superimposing means for superimposing on the path characteristic in synchronization with the frame average pitch, and a filter for approximating the fluctuation component superimposing vocal tract characteristic obtained by the fluctuation component superimposing means by inputting a voiced sound source waveform for each pitch section. A rule synthesizing device comprising: 5. A speech synthesis in which a frame average power, a frame average pitch, and voiced / unvoiced information obtained by speech analysis are input, and a 1-pitch long speech waveform representing a frame in a voiced section determined by voiced / unvoiced information is input. In this method, a frame average power or a frame average pitch is input, and a fluctuation component generating unit that generates a fluctuation component defined by the power or the pitch, and a fluctuation component obtained by the fluctuation generating unit are set to one pitch for each frame average pitch section. A fluctuation component superimposing means for superimposing on the long speech waveform, and a waveform superimposing means for obtaining a synthesized speech by superimposing the fluctuation component superimposing speech waveform of one pitch length obtained by the fluctuation component superimposing means at a frame average pitch interval. A speech synthesis method characterized by 6. A rule synthesizing device for inputting a character string or a phoneme symbol string and outputting synthetic speech according to phonetic or linguistic rules based on dictionary information or phoneme segment information stored in advance, A fluctuation component generating means for generating a fluctuation component defined by a power or a pitch generated based on a rule according to a sentence, and a fluctuation component obtained by the fluctuation generating means for one pitch generated by a rule for each pitch section. A fluctuation component superimposing means for superimposing on a long speech waveform, and a waveform superimposing means for obtaining a synthetic speech by superimposing the fluctuation component superimposing speech waveform of one pitch length obtained by the fluctuation component superimposing means at a pitch interval. A rule synthesizer characterized by.