JPS63110498A - Regular type voice synthesizer - 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 an apparatus for synthesizing speech based on rules from information representing speech such as character strings.

(Prior art) In voice response systems, etc., it may be necessary to synthesize speech not only from specific messages determined in advance during system development, but also from character strings representing arbitrary sentences or word pronunciations during operation. be. In addition, when a machine reads aloud a text intended for human reading, such as a Japanese sentence that includes kanji and kana; it analyzes the text and generates information representing the pronunciation (hereinafter referred to as audio information). Then the audio will be synthesized.

In such cases, you can prepare various speech synthesis rules that control pitch, phoneme duration, amplitude, or spectral parameters, and then apply these speech synthesis rules to the input speech information. So-called rule synthesis of speech is known.

An example of such a rule-based synthesis of speech is given in the Acoustical Society of Japan Speech Study Group Material 585-31 (1985,
7), "Pormant, cv-vc type rule synthesis".

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) obtained by analyzing natural speech in advance. By editing formant parameters and feeding them to a formant-type speech synthesizer, arbitrary speech can be synthesized.

Among these rules, the time length rule is a rule that determines the time length of a phoneme based on the length of a word, the position of a word in a sentence, the position of an accent, and the like.

The pitch rule is a rule that determines the pitch frequency value of each CV-VC based on the length of an exhalation paragraph or clause length or accent, and then determines the pitch of each speech synthesis frame by linearly interpolating them. It is.

The pause rule is a rule that divides a long sentence into several breath paragraphs based on the dependencies between clauses and determines the length of the pause between them.

The parameter editing rule first determines the number of the unit CV-VC voice based on the phoneme sequence of the voice to be synthesized,
This is a rule for synthesizing speech by editing formant data prepared in advance in units of cv-vc.

On the other hand, a second conventional example is "Speech synthesis using articulatory segment editing method", which is presented in Proceedings of the National Conference of IEICE Information Systems Division, No. 1-131, by Mitome, Fushikida, and Takashima. There is a system. Similar to the first conventional example, this synthesizes arbitrary speech by editing unit speech called articulatory segments, but instead of using parameters such as formants, it synthesizes speech waveforms with several pitch levels. This example differs from the first example in that it edits the . Therefore, the time length rules etc. are the same as in the first example, but the difference is that the pitch rules are selected from pitches prepared in advance, and the data is edited taking the pitch into consideration. Become.

In either example, arbitrary Japanese speech can be synthesized based on speech information. Furthermore, many examples are known regarding control rules regarding individual parameters such as time length and pitch.

Furthermore, speech in other languages such as English can be similarly synthesized using rules, and many such examples are known. An example of such is Proceedings IC ANISP 82.
ASSP 82. The Klattalk Text to Speech Conversion System (The Klattalk Text to Speech Conversion System), published on pages 1589 to 1592 of the 1982 International Conference on Acoustics, Speech, and Signal Processing.
-8peech Conversion 5yster
R) There is a J.

This example has a rule that provides target values for various speech synthesis parameters such as pitch, phoneme formant, and amplitude, and generates a time-series pattern for each parameter by smoothly interpolating between them.

What these examples have in common is that various rules are applied to the input audio information, and various parameters (time length, pitch, formant frequency, amplitude target values,
The process involves determining the value of the unit voice (such as the number of the unit voice) and synthesizing the voice based on that value.

(Problem to be solved by the invention) However, in such conventional rule-based speech synthesis devices, the same sentence or word is always pronounced in the same way because the same rule is always applied. It turns out. Therefore, the synthesized speech generated by the conventional regular speech synthesizer has a mechanical and unnatural impression, and it is tiring to listen to it for a long time.

When a word or simple 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 becomes, the more likely this situation will occur, and the more unnatural it will become.

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

(Means for Solving the Problems) Means provided by the present invention to solve the above-mentioned problems is based on the input speech information, and determines the pitch, phoneme duration, amplitude, and spectrum for regular synthesis. A regular speech synthesizer has means for determining the values of speech synthesis parameters such as The apparatus is characterized in that it further comprises means for changing the value of the voice synthesis parameter accordingly.

(Function) The present invention determines the value of the speech synthesis parameter in the same way as in the past using control rules for the speech synthesis parameter, and synthesizes speech by randomly changing the value, thereby preventing the generation of mechanically synthesized speech. It's something you try to avoid. This is because when humans actually read the same sentence aloud, the initial rhythm and timbre vary stochastically each time they are uttered, and the idea is to improve naturalness by realizing this. It is something to do. To this end, depending on the random data generated by the random data generating means,
The means for changing the voice synthesis parameter value changes the parameter value determined in the same way as in the conventional method.

At this time, the actual voice uttered by a human being is analyzed in advance to find out the statistical distribution of the values of the voice synthesis parameters, and the amount by which the voice synthesis parameter values are changed is determined based on the distribution.

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

(Example) Next, the present invention will be explained in detail with reference to the drawings. FIG. 1 is a block diagram of a first embodiment of the invention.

This example is a device that edits unit audio waveforms and synthesizes audio, similar to the second conventional example described above.

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

The unit voice waveform memory 103 stores a plurality of waveform data for each voice, and when a data number assigned to each voice is given from a signal line 118, the data is output from a signal line 119.

Here, numbers of a plurality of waveform data for the same unit sound are consecutively assigned. That is, in the unit speech waveform memory 103, different waveform data for one unit speech are stored as a group. Although the contents of the stored data are different, the configuration of the unit speech waveform memory 103 is similar to that of the second conventional example in which data of the waveform of a unit speech prepared in advance is stored. It is similar to

The data number generation circuit 102 generates a unit speech waveform memory 10 from a series of unit speech names inputted through a signal line 113.
The number of the first data of a group of different waveform data for one unit voice in 3 is generated and output to the signal line 116.

The waveform editing circuit 104 uses only the portion indicated by the time length data input through the signal line 115 out of the waveform data of the unit audio sent from the unit audio waveform memory 103 via the signal line 119, and edits each unit audio waveform. A synthesized speech waveform is generated by interpolating between the two. These data number generation circuit 102 and waveform editing circuit 104 can also be realized with the same configuration as in the second conventional example.

The random number generation circuit 105 generates a random number every time there is 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 waveform data stored for each unit voice.

The adder 106 assigns the random number generation date j? 4)
The random numbers sent from 105 are added and sent to unit audio waveform memory 103 via signal line 118.

When the sequence of unit phonetic names and time length data are input from the signal line 111, the control circuit 101 sends the sequence of unit phonetic names to the data number generation circuit 102 via the signal line 113 to generate a data number. The long data is sent to the waveform editing circuit 104 via the signal line 115.

Furthermore, an instruction is sent to the random number generation circuit 105 via the signal line 114 to generate random numbers. In this way, the random number generator FI inputs the data number generated by the data number generation port 71102.
! The random numbers generated in step 1105 are added, and the data is read out from the unit speech waveform memory 103 and edited in the waveform editing port 104 to generate a synthesized speech waveform, which is output from the signal line 112.

Although this embodiment is described as editing unit speech waveform data, it can also be realized in the same manner as in this embodiment with a device that edits data such as unit speech formand pakun, as in the first conventional example described above. Can be done. That is, instead of waveform data, data such as formant patterns may be stored and edited.

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

Similar to the third conventional example described above, this embodiment provides target values for various speech synthesis parameters such as pitch, phoneme formant, and amplitude, and generates a time series pattern for each parameter by smoothly interpolating between them. It is a type of device that

In the figure, 201 is a control circuit, 202 is a parameter target value generation circuit, 203 is a data interpolation circuit, and 204 is a control circuit.
is a speech synthesis circuit, 205 is a data distribution value memory, 206
is a random number generation circuit, 207 is a multiplier, and 208 is an adder.

Similar to the third conventional example, the parameter target value generation circuit 202 generates a parameter target value based on the phoneme sequence sent from the control port i 201 via the signal line 213, and sends it to the signal line 220. do.

Similarly to the third conventional example, the data interpolation circuit 203 also interpolates between the target values of the parameters and meters input from the signal line 221 based on the control signal sent from the signal line 216, and calculates each parameter. generates a time-series pattern of
Send to 04.

Similarly to the third conventional example, the speech synthesis circuit 204 also generates synthesized speech based on the data sent from the data interpolation circuit 203 and outputs it to the signal line 212.

The data distribution value memory 205 stores the maximum distributed value of the target value of each parameter, and in accordance with the 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. send out.

In the random number generation circuit 206, 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 207, the maximum value of the data distribution sent from the data distribution value memory 205 is multiplied by the random number sent from the random number generation circuit 206, and the product is sent to the adder 208 via the signal line 219.

In the adder 208, the parameter target value generation time H@
The product of the parameter target value sent from the multiplier 202, the data distribution value sent from the multiplier 207, and the random number is added and sent to the signal line 22 as a new parameter target value.
1 to the data interpolation circuit 203.

When the phoneme sequence and time length data are input through the signal line 211, the control port fI@201 sends the phoneme sequence to the parameter target value generation circuit 202 via the signal line 213, generates a parameter target value, and generates a control signal. Signal line 2
16 to the data interpolation circuit 203. Furthermore, an instruction is sent to the data distribution value memory 205 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 206 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 205 and the random number generated by the random number generation circuit 206 is added to the parameter target value generated by the parameter target value generation circuit 202, and the new value is A time series pattern of each speech synthesis parameter is generated based on the parameter target value, and speech is synthesized based on the data.

<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 values of various speech synthesis parameters will be slightly different each time. This has the effect of producing a natural synthesized sound without compromising clarity.

[Brief explanation of the drawing]

1 and 2 are process diagrams showing a first and second embodiment of the present invention, respectively. In the figure, 101 is a control circuit, 102 is a data number generation circuit, 103 is an m-order audio waveform memory, 104 is a waveform editing circuit, 105 is a random number generation circuit, 106 is an adder,
201 is a control circuit, 202 is a parameter target value generation circuit, 203 is a data interpolation circuit, 204 is a speech synthesis circuit, 205 is a data distribution value memory, 206 is a random number generation circuit, 207 is a multiplier, and 208 is an adder. .

Claims (1)

[Claims] means for determining the values of speech synthesis parameters such as pitch, phoneme duration, amplitude, spectrum, etc. for regular synthesis of speech based on input speech information; and means for generating synthesized speech from the values of the parameters. 1. A regular speech synthesis device, further comprising: means for generating random data; and means for changing the value of the speech synthesis parameter according to the value of the random data.