JPS63110497A - Voice spectrum pattern generator - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

Detailed Description of the Invention (Field of Industrial Application) The present invention relates to a speech spectrum pattern used to synthesize speech according to rules from information representing speech such as character strings, that is, a time-varying pattern of the spectrum envelope of speech. The present invention relates to a device for generating.

(Prior art) In so-called speech rule synthesis, in which speech is synthesized from character strings representing the pronunciation of arbitrary sentences or words, speech spectral patterns, that is, phonological It is important to generate time-varying patterns of the spectral envelope of gender-related speech in order to generate clear and natural synthesized speech.

An example of the generation of spectral patterns in the rule synthesis of speech is shown in "Formant, cv-vc type rule synthesis" by Mitome and Fushikida, Acoustical Society of Japan Speech Study Group Material 885-31 (1f185.7). has been done.

This includes voice synthesis rules such as time length rules, pitch rules, pause rules, and parameter editing rules, and is based on CV-VC (C represents a consonant and ■ represents a vowel), which is obtained by analyzing natural speech in advance. By editing the formant parameters and feeding them to a formant-type speech synthesizer, arbitrary speech can be synthesized.

Among these rules, the parameter @ collection rule is based on the phoneme sequence of the speech to be synthesized.
This is a rule for synthesizing speech by editing formant data in units of v-vc to generate a temporal change in formant, that is, a spectral pattern.

In this example, a spectrum pattern for each CV-VC unit is prepared in advance, and by specifying the CV-VC number, a spectrum pattern for that unit part can be generated.

That is, in this case, the number cv'=vc is a parameter for generating a spectrum pattern. Another such example is to use vcv+cvc as the unit, and use PARCOR' instead of formant as the data to be edited.
? Methods using LSP and waveforms of unit speech are known.

On the other hand, as a second conventional example, the procedure
gs ICASSP 82. Proceedings of the 1982 International Conference on Acoustics, Speech and Signal Processing), pages 1589 to 1592
The paper by Klatt shown on page [
The Klattalk Text to Speech Conversion System (The Klattalk T
ext-to-8pee-ch Conversion
system) J is available. This example has a formant rule that gives target values such as formants at the beginning and end of the phoneme of English speech and smoothly interpolates between them. In this case, the target value is a parameter for spectral pattern generation.

What these examples have in common is that various rules that match the conditions are applied to input audio information, parameter values are determined, and audio spectral patterns are generated based on these rules and audio is synthesized.

(Problems to be Solved by the Invention) However, in such conventional speech spectrum pattern generation devices, identical sentences or words, etc.
The same rules always apply, so it will always be pronounced the same way. Therefore, synthesized speech synthesized using spectrum patterns generated by conventional speech spectrum pattern generation devices has a mechanical and unnatural impression, and listening to it for a long time can be tiring.

When a word or a sentence appears as part of a long sentence, if the context is different, it may be synthesized slightly differently depending on the rules. However, when words appear on their own, separated by punctuation marks, and the rules do not take context into account, they will always be combined in the same way.

The longer the text, the more likely this situation will occur, and the more unnatural it will be.

SUMMARY OF THE INVENTION An object of the present invention is to provide a voice spectrum pattern generation device capable of synthesizing more natural speech with a simple configuration such as adding some circuits to a conventional voice spectrum pattern generation device.

(Means for Solving the Problems) Means provided by the present invention to solve the above-mentioned problems is as follows: = 4 - value of parameters for generating a speech spectral pattern based on input speech information and a means for generating a spectrum pattern from the value of the parameter, the apparatus comprising means for generating random data, and a value of the parameter according to the value of the random data. The invention is characterized in that it further comprises means for changing.

(Operation) The present invention determines the value of a parameter for generating a spectral pattern in the same way as before, and randomly changes the value to generate a spectral pattern and synthesize speech, thereby achieving a mechanical method. This is intended to avoid the generation of synthetic speech. This is because when humans actually read the same sentence aloud, the timbre tends to fluctuate stochastically each time they utter it, and by realizing this, the idea is to improve the naturalness. be. To this end, a means for changing parameter values is used to change parameter values determined in the same manner as in the past, and a spectral pattern is generated based on the values, using means for changing parameter values in accordance with random data generated by a means for generating random data. be.

At this time, by analyzing the voice actually uttered by a human in advance to find out the statistical distribution of parameter values related to the spectrum, and determining the amount of change in the parameter value based on that distribution, It is possible to generate more natural synthesized sounds.

As means for generating rank data, methods based on the congruence method and methods based on the M-sequence are conventionally known, and methods based on either method can be used.

(Example) Next, the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram of a first embodiment of the invention.

In the figure, 101 is a control circuit, 102 is a data number generation circuit, 103 is a unit audio data memory, 104 is a data editing circuit, 105 is a random number generation circuit, and 106 is an adder.

The unit audio data memory 103 stores a plurality of spectral pattern data for each unit audio, and when a data number assigned to each is given from the signal line 118, the data is output from the signal line 11g. Here, numbers of a plurality of spectrum pattern data for the same enthronement voice are consecutively assigned. That is, in the unit voice data memory 103, different spectral pattern data for one coronation voice are stored as a group. Although the contents of the stored data are different, the configuration of the coronation voice data memory 103 is similar to that of the first conventional example described above, in which data of the CV-VC spectrum pattern prepared in advance is stored. It is similar to what you keep.

The data number generation circuit 102 generates a sequence of unit voices from the phoneme sequence input from the signal line 113, and generates the first data of a group of different spectral pattern data for one unit voice in the unit voice data memory 103. The data network collection circuit B104 generates a number of 6 and outputs it to the signal line 116.
A spectral pattern is generated by interpolating between each unit voice using only the portion indicated by the time length data input from the signal line 115 out of the spectrum pattern data of the unit voice sent from the signal line 119 through the signal line 119.

These data number generation circuit 102 and data editing circuit 1
04 can also be realized with a configuration similar to that in the first conventional example.

The random number generation circuit 105 generates a random number every time it receives an instruction from the control circuit 101 and sends it to the adder 106 via a signal line 117.
send to The random number generated at this time takes a value from 0 to (N-1>), where N is the number of spectrum pattern data stored for each unit voice.

The adder 106 adds the random number sent from the random number generation circuit 105 to the number of the first data of the group of different spectral pattern data for one unit voice sent from the data number generation circuit 102, and adds the random number sent from the random number generation circuit 105 to the signal line 118. The data is sent to the unit audio data memory 103 via the audio data memory 103.

When the control circuit 101 receives the phoneme sequence and time length data from the signal line 111, it sends the phoneme sequence to the data number generation circuit 102 via the signal line 113 to generate a data number, and sends the time length data to the signal line 115. The data is sent to the data editing circuit 104 via. Furthermore, an instruction is sent to the random number generation circuit 105 via the signal line 114 to generate random numbers. thus,
In the data number generated by the data number generation circuit 102,
The random numbers generated by the random number generation circuit 105 are added, and the data is read out from the unit audio data memory 103 and edited in the data editing circuit 104 to generate a spectrum pattern, which is output from the signal line 112.

FIG. 2 is a block diagram of a second embodiment of the invention.

In the figure, 201 is a control circuit, 202 is a polar target value generation circuit, 203 is a data interpolation circuit, and 204 is a control circuit.
is a data distribution value memory, 205 is a random number generation circuit, 206
is a multiplier, and 207 is an adder.

The formant target value generation circuit B2O2 generates a formant target value based on the phoneme sequence sent from the control circuit 201 via the signal line 213, and sends it to the signal line 220, as in the second conventional example described above. Send.

Similarly to the second conventional example, the data interpolation circuit 203 also generates a spectrum pattern by interpolating between the formant target values input from the signal line 221 based on the time length data sent from the signal line 216. do.

The data distribution value memory 204 stores the maximum distributed value of the target value of each formant data, and according to an instruction sent from the control circuit 201 via the signal line 214, the maximum value of the distribution is sent to the signal line 217. and send it off.

In the random number generation circuit 205, the signal line 2 is connected from the control circuit 201.
15, positive and negative random numbers having an absolute value of 1 or less are generated and sent to the signal line 218.

In the multiplier 206, the maximum value of the data distribution sent from the data distribution value memory 204 is multiplied by the random number sent from the random number generation circuit 205, and the product is sent to the adder 207 via the signal line 219.

In the adder 207, the formant target value generation circuit 2
The formant target value sent from 02 and the product of 1, the data distribution value sent from multiplier 206, and the random number are added together and sent to signal line 22 as a new formant target value.
1 to the data interpolation circuit 203.

When a phoneme sequence and time length data are input from a signal line 211, the control circuit 201 sends the phoneme sequence to a formant target value generation circuit 202 via a signal line 213 to generate a formant target value, and converts the time length data into a signal. line 2
16 to the data interpolation circuit 203. Furthermore, an instruction is sent to the data distribution value memory 204 via the signal line 214 to generate the maximum value of the data distribution, and an instruction is sent to the random number generation circuit 205 via the signal line 215 to generate random numbers.

In this way, the product of the maximum value of the data distribution generated by the data distribution value memory 204 and the random number generated by the random number generation circuit 205 is added to the formant target value generated by the formant target value generation circuit 202, and the new value is generated. A spectral pattern is generated based on the formant target value and time length data, and is output from the signal line 212.

(Effects of the Invention) As explained above, according to the present invention, even if the same expression repeatedly appears in the speech information to be synthesized, the parameter values take slightly different values each time, making it easy to understand clearly. This has the effect of producing a natural synthesized sound without compromising gender.

[Brief explanation of the drawing]

FIGS. 1 and 2 are block diagrams showing first and second embodiments of the present invention, respectively. In the figure, 101 is a control circuit, 102 is a data number generation circuit, 103 is a unit audio data memory, 104 is a data editing circuit, 105 is a random number generation circuit, 106 is an adder,
201 is a control circuit, 202 is a formant target value generation circuit, 203 is a data interpolation circuit, 204 is a data distribution value memory, 205 is a random number generation circuit, 206 is a multiplier, 2
07 is an adder.

Claims (1)

[Claims] An audio spectral pattern generation device comprising: means for determining the value of a parameter for generating an audio spectral pattern based on input audio information; and means for generating the spectral pattern from the value of the parameter. and means for changing the value of the parameter according to the value of the random data.