JPS61290499A - Enunciation speed varying voice synthesizer - Google Patents

Abstract

(57) [Summary] 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 synthesis device with variable speaking rate.

(Prior art and its problems) Conventionally, a speech synthesizer with a variable speaking rate has a sound source circuit that generates a sound source waveform based on sound source information such as pitch, voiced or unvoiced, and a sound source circuit that generates a sound source waveform based on spectral information. and a programmable calculation counter, and each time the time counter counts a preset time, the sound source waveform generated by the sound source circuit is applied to the synthesis filter while updating the audio information. There is a known device for synthesizing speech by filtering, which changes the speaking speed by changing the time data set in the time counter. That is, for example, when synthesizing speech using speech information obtained by analyzing natural speech waveforms at regular intervals,
If time data is set in the time counter so that the voice information is updated at a time different from the time of analysis, the overall t-voice time can be changed, and therefore the speaking speed can be changed. For example, if speech is synthesized using data analyzed every 10 milliseconds and updated every 8 milliseconds, speech can be produced 25% faster. An example of such a speech synthesizer is NE
This is shown in "Speech synthesis LSI J using formants as parameters" by Suzuki et al. published in C Technical Report Vol. 1.34 No. 10 p. 92-97.

As a second conventional example, a method is known in which regular synthesis for synthesizing speech from character strings involves expanding and contracting constant parts of speech (for example, constant parts of vowels). In other words, when synthesizing arbitrary speech by editing pre-stored speech information data in units of monosyllables, ox-cart syllables, etc., each unit is This changes the duration of the audio. This is originally
Although it is intended to implement rules for synthesizing speech with natural rhythms, it can also be used to change the rate of speech as a result. An example of this is the Journal of the Institute of Electronics and Communication Engineers Vol.
1-D No., 11 p., 858-865, "Speech Synthesis Method Using PARCOR-VCV Chain" by Sato.

According to these conventional examples, it is possible to change the speaking speed of the voice without changing the pitch or pitch of the voice with a simple structure, but each of them has the following drawbacks.

In the first conventional example, in principle, it is possible to utter as fast as you like as long as the data update control can keep up, but with consonants such as plosives, the intelligibility decreases as the duration becomes shorter, so I couldn't do it fast. Also,
In the second conventional example, if consonant parts with short durations are not thinned out, the intelligibility will not deteriorate as in the first conventional example, but the time length of the constant part that can be thinned out will not deteriorate. It was not possible to increase the speed very much because of the limitations.

Therefore, for example, in a device that is required to speak at twice the normal speed in order to grasp the gist of a sentence, the above-mentioned conventional example could not achieve sufficient performance.

(Objective of the Invention) An object of the present invention is to provide a variable speech rate speech synthesis device that is capable of high-speed speech production, which has not been possible with conventional examples, without significantly reducing the intelligibility.

(Structure of the Invention) According to the present invention, there is provided a means for storing sound information including sound source information and spectral information, a sound source circuit that generates a sound source waveform based on the sound source information, and a frequency characteristic based on the spectral information. has a synthesis filter that is controlled, and a programmable time counter, and each time the time counter counts a preset time,
In an apparatus for synthesizing speech by filtering a sound source waveform generated by the sound source circuit with the synthesis filter while updating the sound information, a stationary portion of the sound of the sound information is Along with thinning out the
There is obtained a speech synthesis device with a variable speaking rate, characterized in that it includes means for controlling to change the time data set in the time counter.

(Principle of the Invention) The present invention improves the rate of speech by shortening the time length of the stationary part to the extent that phonology is not lost, and by shortening the update time of speech information to the extent that the intelligibility of plosives etc. does not deteriorate. change. Thereby, high-speed speech production can be realized without significantly reducing the clarity.

(Example) Next, the present invention will be explained in detail using the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention. In the figure, 101 is a control circuit, 102 is a programmable time counter, 103 is a sound source circuit, 104 is a synthesis filter, 105 is an audio information data memory, 106 is an address table, and 107 is a time length memory.

The voice information data memory 105 stores voice information of voice messages, that is, information data consisting of sound source information and spectrum information. Further, the address table 106 stores the start address of data corresponding to each voice message. Further, the time length memory 107 alternately stores the time lengths of non-regular portions and regular portions of each voice message.

First, when control information consisting of a message number, a data update time value, and an expansion/contraction rate value of the stationary section is input to the control circuit 101 from the input terminal 108, the control circuit 1
01 indicates the message number in the address table 106.
and signal lines 110 and 1 to the time length memory 107, respectively.
19, information on each address is read out via a signal line 111, and time length information is read out via a signal line 120.

Next, the control circuit 101 sends the address value to the audio information data memory 105 via the signal line 112 while sequentially updating the address value starting from the value of the first address, and transfers the audio information data to the signal line. 113 sequentially. At this time, the time length data of the steady part is multiplied by the expansion/contraction rate of the steady part of the control information, only the data of the steady part is read out, and the next data is read from the beginning of the next unsteady part. Start reading.

On the other hand, the control circuit 101 sends the update time value of the data to the programmable time counter 102 via the signal line 114. The programmable time counter 1
02 sends timing information to the control circuit 101 via the signal line 115 every time a given update time elapses. In synchronization with this timing information, the control circuit 1
01 is the signal line 1 for the sound source information of the audio information data.
16 to the sound source circuit 103, and spectrum information is sent to the synthesis filter 104 via a signal line 117, respectively.

The sound source circuit 103 generates a sound source waveform such as a random sound source waveform or a periodic sound source waveform according to the sound source information, and sends it to the synthesis filter 104 via a signal line 118.

The synthesis filter 104 filters the sound source waveform sent from the sound source circuit 103 based on the spectrum information and sends it to the output terminal 109 as synthesized speech.

According to this embodiment, the speaking speed can be varied over a wide range by externally providing the data update time value and the expansion/contraction rate value of the stationary portion.

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

This embodiment is an example in which the present invention is applied to speech synthesis based on rules, and arbitrary speech is synthesized by editing unit speech data such as a single speech or a bullock-cart syllable. Processing of rules regarding accents and prosody will be performed by another device, and the input information to this device includes the unit voice number, expansion/contraction information of the stationary part, interpolation control information between unit voices, and data update time information. , pitch information, etc.

In FIG. 2, reference numeral 202 denotes a unit voice data memory, which stores voice information data such as sound source information and spectrum information of unit voices such as monosyllables and oxcart syllables. Reference numeral 203 denotes an address table, which stores the start address of each unit audio data, the addresses of the regular part, the non-regular part, and the time length of each part.

First, when the input information is input from the input terminal 209 to the editing control circuit 201, the editing control circuit 201 sends the number of each unit audio to the address table 203 via the signal line 211, and start address,
The addresses of the steady portion and unsteady portion, the time length of each portion, etc. are read out via the signal line 212. Next, the editing control circuit 201 sends the address data to the unit audio data memory 202 via the signal line 213, and reads the audio information data via the signal line 214. At this time, the editing control circuit 201 reads out only necessary data for the stationary part based on the expansion/contraction information of the stationary part of the input information, and reads out only the necessary data based on the interpolation control information for between each unit voice. Generate data for the interpolation section. Further, the editing control circuit 201 edits these data, the data update time information and pitch information of the input information, and while checking the empty state of the buffer memory 204 via the signal line 216 The edited audio information data is sent to the buffer memory 204 via line 215.

The data update control circuit 205 sends a read signal to the buffer memory 204 via a signal line 218, sequentially reads out the edited audio information data via a signal line 217, and 'data of the data is The update time value is sent to the programmable time counter 20 via the signal line 219.
Send to 6.

The programmable time counter 206 transmits timing information to the signal line 2 every time a given update time elapses.
20 to the data update control circuit 205.

In synchronization with this timing information, the data update control circuit 205 sends the sound source information of the audio information read out from the buffer memory 204 to the sound source circuit 207 via the signal line 221 and spectra information to the sound source circuit 207 via the signal line 222. and sent to the synthesis filter 208, respectively.

The sound source circuit 207 generates a sound source waveform such as a random sound source waveform or a periodic sound source waveform according to the sound source information, and sends it to the synthesis filter 208 via a signal line 223.

The synthesis filter 208 filters the sound source waveform sent from the sound source circuit 207 based on the spectrum information, and sends the filtered sound to the output terminal 210 as synthesized speech.

According to this embodiment, the audio of any text can be synthesized at nearly double the normal speaking speed, so if it is combined with a device that analyzes text, for example, it is possible to understand the outline of a large amount of documents in a short time. Become.

(Effects of the Invention) As described above, according to the present invention, high-speed speech production can be realized without significantly reducing the clarity. Furthermore, since the degree of freedom in controlling temporal changes in the spectrum is greater than in the conventional method, there is an effect that more natural speech may be synthesized in speech synthesis based on rules.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG. 2 is a block diagram of an embodiment of the present invention.
FIG. 3 is a block diagram of a second embodiment of the present invention. In the figure, 101 is a control circuit, 102 is a programmable time counter, 103 is a sound source circuit, 104 is a synthesis filter, and 10
5 is an audio information data memory, 106 is an address table, 107 is a time length memory, 201 is an editing control circuit, 20
2 represents a unit audio data memory, 203 an address table, 204 a buffer memory, 205 a data update control circuit, 206 a programmable time counter, 207 a sound source circuit, and 208 a synthesis filter. Substitute, buried soil, inner thickness 2r diagram

Claims (1)

[Claims] means for storing audio information consisting of sound source information and spectral information; a sound source circuit that generates a sound source waveform based on the sound source information; a synthesis filter whose frequency characteristics are controlled based on the spectral information; and a programmable timer. a counter, each time the time counter counts a preset time, the sound information is updated, and the sound source waveform generated by the sound source circuit is filtered by the synthesis filter. A device for synthesizing speech, characterized by comprising means for thinning out a steady portion of the speech from the speech information and controlling to change time data set in the time counter based on instructions from the outside. A speech synthesizer with variable speaking speed.