JP3083624B2 - Voice rule synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech rule synthesizing apparatus suitable for synthesizing speeches having different voice qualities.

[0002]

2. Description of the Related Art A speech rule synthesizing method uses a speech parameter obtained by observing or analyzing a voice actually uttered by a human being as a consonant + vowel (CV), a vowel + consonant + vowel (VCV), a consonant + vowel. A voice unit is prepared in advance in units of + consonants (CVC) or the like, and these voice units are interpolated and connected in accordance with the input phoneme sequence. From the phoneme parameters thus obtained and the pitch pattern generated on the other hand, Is sent to the synthesizer to synthesize speech.

In order to synthesize voices having different voice qualities by the rule synthesis method described above, all of the voice segments are created from voice parameters obtained from voice data uttered by different people. Switching to speech units may be used.

[0004]

As described above, when synthesizing voices having different voice qualities by using the conventional voice rule synthesizing technique, all voice segments must be replaced, and one voice unit must be replaced. When synthesizing voices of a plurality of voice qualities with a synthesizer, there is a problem that the memory required for holding speech units increases in proportion to the number of voice qualities to be synthesized.

Therefore, the present invention can change the voice quality only by exchanging a small number of voice segments, and is necessary for holding voice segments even when it is desired to synthesize voices of a plurality of voice qualities with one synthesizer. It is an object of the present invention to provide a speech rule synthesizing apparatus that requires only a small amount of memory.

[0006]

Means for Solving the Problems] speech synthesis by rule instrumentation of the present invention
In order to solve the above problem , multiple
Vowels consisting of vowels and multiple semi-vowel speech parameters
Vowel segment storage unit that stores segment and semi-vowel segments, and specific voice
Speech parameters of consonants including the middle part of quality vowels
Segment storage unit for storing a consonant segment composed of data
Vowel segments and semi-vowel segments stored in the phoneme segment storage unit
Multiple basic pitch information according to the voice
Basic pitch storage unit to remember and input to be synthesized
Analyzes documents and generates phoneme series and accent information
And a language analysis and phoneme length determining unit for determining phoneme length.
Based on the phoneme sequence generated by the language analysis and phoneme length determination unit
Vowel segments are input and specified as voice quality specification information.
Vowel segments and semi-vowel segments corresponding to the selected voice quality
Read from the unit storage unit, and the above
Read from the segment storage unit and perform the above language analysis and phoneme length
Interpolation connection according to the phoneme length determined by the determination unit
A phonological parameter generator for generating parameters, the language
Accent information and decision generated by the analysis / phoneme length determination unit
Phoneme length specified and input as voice quality specification information
Pitch putter based on basic pitch information corresponding to voice quality
Parameter generation to generate prosody parameters consisting of
And the prosody parameter generated by the prosody parameter generation unit.
Generates a sound source pulse based on the parameters
The phoneme parameters generated by the data generator
And a synthesis filter unit that performs voice synthesis .

[0007]

According to the above arrangement , the speech rule of the speech unit which reads out the speech parameters of the speech segments stored in the storage means in advance in accordance with the input phoneme sequence and sequentially connects them is given to the speech synthesizer to output the speech. in the device, for the voice segment other than a vowel segment and glide piece, always speech unit is used consisting of the same speech parameters of voice, for the vowel segment and glide element pieces of different vowels and semi-vowels of voice quality Vowel segments and semi- vowel segments consisting of speech parameters are switched and used.

The vowel part including the vowel segments is the part that most affects the voice quality of the voice. Therefore, by switching and using a vowel segment composed of vowel speech parameters having different voice qualities at least as the vowel segment, other vowel segments (for example, speech segments other than vowel segments and semi-vowel segments) are used. In this case, it is possible to switch the voice quality of the synthesized voice even if a voice segment consisting of voice parameters having the same voice quality is always used.

That is, according to the speech rule synthesizing apparatus of the present invention, it is possible to change the voice quality of synthesized speech only by replacing a small number of speech units, and to synthesize voices of a plurality of voice qualities with one synthesizing apparatus. Even in this case, a small increase in the memory required for holding the speech unit is required.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a speech synthesizer according to one embodiment of the present invention.
FIG. 3 is a block diagram of a (voice rule synthesis device) .

The speech synthesizing apparatus shown in FIG.
For example, it has three vowel unit storage units 2 to 4 and three basic pitch storage units 12 to 14. The unit storage unit 1 stores a consonant segment Cv including speech parameters of a consonant including a part of the way to a vowel and a vowel transition unit segment vc at the time of transition from a vowel to a consonant. The vowel segment storage units 2 to 4 store vowel segments each including voice parameters of vowels having different voice qualities. The basic pitch storage units 12 to 14 store the basic pitches of the speakers of the voices that are the basis of the vowel segments stored in the vowel unit storage units 2 to 4, respectively.

The speech synthesizing apparatus shown in FIG. 1 also generates a phonological sequence based on linguistic analysis of a text (input document) to be subjected to voice synthesis (speech), generates accent information, and determines each phonological length. It has a phoneme length determination unit 5 and a control unit 6 that controls a phoneme parameter generation unit 7 and a prosody parameter generation unit 8 described below.

The speech synthesizing apparatus shown in FIG. 1 also uses the phoneme sequence and phoneme length generated by the language
Phoneme parameter generation for generating phoneme parameters using a speech unit stored in the unit storage unit 1 and a speech unit stored in a selected one of the vowel unit storage units 2 to 4 It has a unit 7, a prosody parameter generation unit 8, and a synthesizer filter 9.

The prosody parameter generation unit 8 stores the accent information and phoneme length generated by the language analysis and phoneme length determination unit 5 and the basic pitch stored in one of the basic pitch storage units 12 to 14. And a prosody parameter is generated based on The synthesizer filter 9 generates a synthesized speech based on the phoneme parameters from the phoneme parameter generation unit 7 and the prosody parameters from the prosody parameter generation unit 8.

The voice synthesizing apparatus of FIG. 1 further includes a vowel unit switching unit 21 for selecting one of the vowel unit storage units 2 to 4 and switching connection to the phoneme parameter generation unit 7, and a basic pitch storage unit 1.
A basic pitch switching unit 22 for selecting one of 2-14 and switching connection to the prosody parameter generation unit 8;
And 3. The voice quality switching control unit 23 performs a vowel unit switching unit 21 and a basic pitch switching unit 22 according to voice quality specification information (user specification) input from an input unit (not shown).
, The switching of the voice quality of the voice synthesized by the synthesizer filter 9 is controlled.

Next, the speech synthesis processing in the speech synthesis apparatus shown in FIG. 1 will be described. First, a method of creating speech units stored in the unit storage unit 1 and the vowel unit storage units 2 to 4 will be described in detail.

In preparing a speech unit, first, a time window having a fixed time length of about 20 msec is applied to voice data uttered by a specific speaker such as an announcer in accordance with the utterance list, and shifted by a fixed time of about 10 msec. Cepstrum analysis is performed in each window.

Next, while observing the power spectrum and audio power of each frame, cepstrum parameters corresponding to a frame range to be cut out as a segment are extracted.
Speech unit. FIG. 2A shows an example in which Cv and vc segments are cut out from one phoneme of audio data. As described above, the transition section is cut out in a relatively wide range (the number of frames is large). The Cv and vc segments are stored in the segment storage unit 1 as described above.

On the other hand, only a cepstrum parameter for one frame is cut out from a vowel part which is a stationary section. here,
Each vowel in the voice data of three speakers (for Japanese,
Cepstrum parameters are cut out from / a /, / i /, / u /, / e /, / o /) as described above, and stored in the vowel unit storage units 2 to 4, respectively. In this case, the basic pitches of the three speakers are stored in the basic pitch storage units 12 to 14.

The input document to be subjected to speech synthesis (speech) is provided to a language analysis / phoneme length determination unit 5. The linguistic analysis / phoneme length determination unit 5 performs linguistic analysis on the input document to generate a phonological sequence and accent information,
Then, each phoneme length is determined, and the information is passed to the control unit 6. The control unit 6 controls the phoneme parameter generation unit 7 and the prosody parameter generation unit 8 according to the information from the language analysis and phoneme length determination unit 5.

On the other hand, the voice quality switching control unit 23 is supplied with voice quality specifying information indicating the voice quality of the synthesized voice specified by the user through an input unit (not shown). Voice quality switching control unit 2
3 controls the vowel unit switching unit 21 and the basic pitch switching unit 22 according to the voice quality designation information.

Thus, the vowel unit switching unit 21 selects the vowel unit storage unit corresponding to the voice quality indicated by the voice quality designation information from the vowel unit storage units 2 to 4, and selects the phonological parameter generation unit 7.
Connect to Further, the basic pitch switching unit 22 selects a basic pitch storage unit corresponding to the voice quality indicated by the voice quality designation information from the basic pitch storage units 12 to 14, and connects the selected basic pitch storage unit to the prosody parameter generation unit 8. Here, the vowel unit storage unit 2 and the basic pitch storage unit 12, the vowel unit storage unit 3 and the basic pitch storage unit 13, or the vowel unit storage unit 4 and the basic pitch storage unit 14
Is selected in combination.

The phoneme parameter generation unit 7 receives a control from the control unit 6 and, based on the phoneme sequence (reading phoneme sequence) generated by the language analysis and phoneme length determination unit 5, extracts necessary speech units. A vowel segment composed of vowel speech parameters is one of the vowel segment storage units 2 to 4 switched by the vowel unit switching unit 21 (that is, a vowel unit storage unit corresponding to a specified voice quality). ), The consonant segments (here, the Cv segment and the vc segment) composed of the consonant voice parameters are read from the segment storage unit 1 and determined by the language analysis / phoneme length determination unit 5. Interpolation connection is performed according to each phoneme length to generate phoneme parameters.

The connection between the segments is made as shown in FIG. That is, by inserting and adjusting the repetition section and the interpolation section of the frame, the segments are connected with each other while adjusting to the respective phoneme lengths determined by the language analysis / phoneme length determination section 5. Here, linear interpolation of each order of the cepstrum parameter is used as an interpolation method.

The prosody parameter generation unit 8 receives control from the control unit 6, and switches the accent information and the phoneme length generated by the language analysis / phoneme length determination unit 5 and the basic pitch switched by the basic pitch switching unit 22. Storage unit 12-14
Is generated based on the basic pitch (that is, the basic pitch corresponding to the designated voice quality) stored in one of them.

The prosody parameters generated by the prosody parameter generation unit 8 are supplied to a synthesizer filter 9. The synthesizer filter 9 is also supplied with the phoneme parameters generated by the phoneme parameter generator 7. The synthesizer filter 9 is, for example, an LMA filter (logarithmic amplitude characteristic approximation filter), generates a sound source pulse based on the prosody parameters from the prosody parameter generation unit 8, and converts the phoneme parameters from the phoneme parameter generation unit 7 into filter coefficients. To produce synthesized speech.

In the speech synthesized in this way, only the vowel part is selected from the selected vowel segment (ie, the vowel segment read out from the vowel segment storage unit selected by the vowel unit switching unit 21). The resulting voice quality. However, since the vowel part is the part that most affects the voice quality of the voice, the voice synthesized as described above becomes the voice quality of the speaker who is the source of the vowel unit specified by the user.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment. For example, in the above embodiment, only the vowel segments are switched to vowel segments created from speech parameters having different voice qualities to change the voice quality, but a semi-vowel (/ w /, / y / Etc.) also have a relatively large effect on the voice quality of the synthesized voice, so that these semi-vowel segments are provided independently for each voice quality, like the vowel segments, and switched according to the voice quality specified by the user. You can also.

Further, in order to further improve the quality of the synthesized speech, the sound source residual signal of each vowel is held together for each vowel unit having a different voice quality, and based on this, the synthesizer 9 uses this signal. It is also possible to generate a sound source and synthesize speech.

In the above embodiment, the vowel segments having different voice qualities (vowel segment storage units 2 to 4 storing the vowel segments) correspond to different vowel segments. Although the basic pitch storage units 12 to 14 storing the basic pitch of the speaker of the voice are provided, only the basic pitch of the specific speaker may be prepared and used in common. Absent. However, in this case, the pitch may not be the voice pitch of the voice data speaker that is the source of the vowel unit.

Further, there is no limitation on the type of the synthesis parameters or the speech unit connection method, and other parameters such as LPC (Linear Predictive Coding) other than the cepstrum parameters may be used. In short, the present invention can be variously modified and implemented without departing from the gist thereof.

[0032]

As described above in detail, according to the present invention, according to the present onset Akira,
Can change the voice quality by simply interchanging a small number of speech units, also even if you want to synthesize speech of a plurality of voice in one combined unit, requires only a small amount is increased memory required to hold the speech unit .

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech synthesizer according to an embodiment of the present invention.

FIG. 2 is a view for explaining extraction of speech units and connection of speech units.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Unit storage part, 2-4 ... Vowel unit storage part, 5 ... Language analysis and phoneme length determination part, 6 ... Control part, 7 ... Phoneme parameter generation part, 8 ... Prosody parameter generation part, 9 ... Synthesizer Filters, 12 to 14: Basic pitch storage unit, 21: Vowel unit switching unit, 22: Basic pitch switching unit, 23: Voice quality switching control unit.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-59-72494 (JP, A) JP-A-60-115997 (JP, A) JP-A-62-229199 (JP, A) JP-A 61-72199 166600 (JP, A) JP-A-1-271800 (JP, A) JP-A-64-35598 (JP, A) JP-A-2-205896 (JP, A) Patent 2703253 (JP, B2) Supervised by Oizumi, Fujimura “Speech Science” (1972-3-10), University of Tokyo Press, p. 349-364 Furui "Digital Speech Processing" (1985-9-25) Tokai University Press p. 141-142 (58) Field surveyed (Int.Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00-21/06

Claims (1)

(57) [Claims] A plurality of vowels and a plurality of halves for different voice qualities.
Vowel segments and semi-vowel segments consisting of vowel speech parameters
Vowel segment storage unit that stores vowels, and consonant voices that include the middle of a vowel with a specific voice quality
A unit storage unit for storing a consonant unit composed of parameters, and a vowel unit and a semi-vowel stored in the vowel unit storage unit
Basic pitch information corresponding to the voice quality on which the phoneme was based
A basic pitch storage unit that stores a plurality of data and an input document that is a target of speech synthesis
A linguistic solution that generates sequence and accent information and determines phoneme length
And analysis and phoneme length determining unit, also the language phoneme sequence generated by the analysis-phoneme length determining unit
Voice quality designation information for vowel segments and semi-vowel segments
Vowel segments and semi-vowels corresponding to voice quality specified as input
A phoneme segment is read from the vowel segment storage unit,
For the piece, read from the segment storage unit,
Complement according to the phoneme length determined by the language analysis and phoneme length determination unit.
Phonological parameter generation that generates phonological parameters by interconnecting
The accent part generated by the linguistic analysis and phoneme length determining part.
Information and the determined phoneme length, and
Based on the basic pitch information corresponding to the specified voice quality
Prosody that generates prosodic parameters consisting of pitch patterns
A parameter generation unit, and a prosody parameter generated by the prosody parameter generation unit.
Generating a sound source pulse based on
Using the phoneme parameters generated by the
And characterized by including a synthesis filter unit for performing voice synthesis
Voice rule synthesis apparatus.