JPS60500A - Speech analysis and synthesis method - 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 digitizes an audio signal, analyzes it, transmits or stores the parameters obtained by the analysis at a low data rate, and restores the audio signal again. This invention relates to a speech analysis and synthesis device. - Conventional configuration and its problems Normally, in a speech analysis and synthesis device, vocal tract parameters and sound source parameters are extracted from input speech using an analyzer.
Each parameter is subjected to processing to reduce the data rate, such as encoding, and sent to a transmission line or storage element, and then resynthesized into speech by a synthesizer. Conventionally, the method in this case is
(1) A method in which the analysis residual waveform extracted at the time of analysis is transmitted or stored as it is, or compressed by processing such as a difference, and (2) an amplitude parameter representing the loudness of the sound; Pitch, which indicates the pitch of the voice
(3) In the example of storing voice data, each method of the above 2 (method of extracting and transmitting or storing the parameter and voiced/unvoiced switching information) There is a method that resynthesizes the speaker's voice quality well, but the disadvantage is that the data rate during transmission or storage is high. On the other hand, although the data rate is low, it uses voiced sound source data that is constant regardless of the speaker, so it has the disadvantage of resulting in a synthesized sound that loses individual characteristics such as the luster of the voice.

(Method '4' has intermediate characteristics between methods (1) and (b), but is not suitable for real-time analysis and synthesis in which the speaker is not constant.

OBJECT OF THE INVENTION The present invention aims to improve the above-mentioned drawbacks of the prior art, and its purpose is to reproduce speech that includes the individual characteristics of the speaker without significantly increasing the amount of information in real-time transmission of speech. This provides a transmission method that does not involve combining.

Structure of the Invention The present invention includes a parameter analysis section that converts speech into a digital signal and extracts each parameter of the vocal tract, amplitude, pitch, and voiced/unvoiced judgment using a linear prediction method, etc., as well as a parameter analysis section that converts speech into a digital signal and extracts each parameter of vocal tract, amplitude, pitch, and voiced/unvoiced judgment. The analyzer has a circuit that extracts it as sound source data,
It has a transmission line that transmits parameters and voiced driving sound source data at a low data rate, and uses a predetermined waveform before the voiced driving sound source data is transmitted, and a new waveform after the voiced driving sound source data is transmitted. This is a speech analysis and synthesis device having a synthesizer that synthesizes speech using driving sound source data.Description of EmbodimentsHereinafter, embodiments of the present invention will be described in detail.0Figure 1 shows a speech analysis and synthesis device according to the present invention. In FIG. 1, which is a block diagram showing the configuration, numeral 1 collects the transmitted sound with a sound collector such as a microphone, converts it into an analog signal, and supplies it to the sound analyzer 2. Speech analyzer 2 converts the analog signal into 8~
After sampling at approximately 10 KHz and converting to a digital signal, vocal tract parameters and sound source parameters are determined by linear predictive analysis for each interval of approximately 6 to 20 ms (called a frame), and these parameters are used for encoding, etc. The band is further compressed and sent to the transmission line 3.The transmission line 3 is not only a system that transmits in real time like a normal telephone line, but also a storage system such as a writable memory element (RAM). The speech synthesizer 4 that receives the zero compression parameter, which may be a medium, performs the inverse operation of the band compression performed by the speech analyzer 2 to restore the speech signal. The restored audio signal is given to the speaker 6 for audio reproduction.

As a band compression technique, this embodiment uses the PAftCOR method, which is one of the linear predictive analysis methods.

A speech analyzer using the PARCOR method will be described later.

When performing PARCOR analysis, parameter transmission, and PARCOR synthesis of the audio signal shown in Figure 2, the capacity of the transmission path is determined by the maximum data rate per unit time of the parameters.
In actual transmission, the transfer data rate is extremely low in silent sections such as sections (1) 9 (3) 9 (5) compared to sections (→, (4)) in FIG. Of the residual waveforms obtained by analyzing sections (2) and (4), a part of the stationary vowel section is transmitted in sections 0) and (phrases) and used as a voiced driving sound source in the synthesizer. Since the difference waveform includes the individuality of the speaker that is not expressed by parameters, speech with rich individuality can be synthesized.Voiced driving sound source data is normally a data string of one pitch period or less, but in this example, it is 8 pitch periods or less. Since it is composed of bits x 31 points, it is necessary to transfer 248 bits to the silent section, and if a 2400 bit/second transmission line is used, it will take approximately 100 milliseconds to transmit this residual data. However, in normal vocalization, it is several hundred mi+7
There are often silent intervals of about seconds, so it can be transmitted sufficiently.

Note that voiced driving sound source data can be compressed using a differential method or the like and transmitted in a short silent section.On the other hand, the synthesizer does not transmit driving sound source data yet at the start of vocalization, such as in section (2). Therefore, a predetermined inno<lus waveform or the like is used as the voiced drive sound source data.0 FIG. 3 is a block A showing the configuration of a speech analyzer corresponding to 2 in FIG. 1. 21 is a mu-D converter that converts the audio signal into a sampled digital signal; the digital signal is sent to the PA, the ROOR analyzer 22, the pitch extractor 23, the voiced/unvoiced determiner 24, and the silent section detector 26;
A part of the residual signal obtained by the RCOf'1 analyzer 22 is extracted in a residual extraction circuit 26 and temporarily stored for viewing.An amplitude parameter is determined in an amplitude determination circuit 27.

The parameters determined by the PARCOFT analyzer 22, pitch extractor 23, voiced/unvoiced determiner 24, silent section detector 26, and amplitude determination circuit 27 are encoded by an encoder 28,
It is sent to the transmission line 3 via the switch 29 as a parameter value representative of a certain time interval (frame). When the silent interval detector 25 detects a silent interval, the switch 29 is inverted, and a portion of the residual waveform extracted by the residual extraction circuit 26 is transmitted as voiced sound source data to the synthesizer.

FIG. 4 is a block diagram showing the configuration of a PARCOR type speech synthesizer that receives transmitted parameters and voiced sound source data and synthesizes a speech signal, and corresponds to 4 in FIG. 1. The transmitted data is separated into two types by a selector 41, parameters are stored in a parameter memory 42, and voiced sound source data is stored in a sound source memory 43.

Immediately after the power is turned on or when a long period of silence is detected, voiced sound source data such as the predetermined impulse waveform is automatically set in the sound source memory 43 until new voiced sound source data is set by the selector 41. Retained. 4
4 is a voiceless sound source generator and is voiced.

The data of either the sound source memory 43 or the unvoiced sound source generator 44 is selected by the silent selector 46, and the data in the parameter memory 42 is calculated by the digital filter 46. The signal is converted into an audio signal, amplified by an amplifier 48, and supplied to the speaker 6.

Effects of the Invention As described above, the present invention analyzes a stationary voiced sound section when analyzing, transmitting, and synthesizing speech waveforms in real time, and uses a portion of the residual waveform obtained as voiced driving sound source data to generate silent sound. By transmitting a part of the residual waveform extracted during speech analysis to the silent section where the parameter is not transmitted, the speech analysis and synthesis device transmits data to the middle low data rate section.
It is possible to synthesize speech based on the voice quality layer 75 unique to the speaker, such as the luster and roundness of the voice, without increasing the maximum transfer data capacity of the transmission path.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of the speech analysis and synthesis device according to the present invention, FIG. 2 is a waveform diagram showing the time relationship between speech waveforms and transmitted data, and FIG. 3 is the speech analysis and synthesis device according to the invention. A block diagram showing the structure of the analyzer section. FIG. 4 is a block diagram showing the structure of the speech synthesizer section of the speech synthesis and analysis apparatus according to the present invention. 1...Sound collector, 2...Speech analyzer, 3
...Transmission line, 4...Speech synthesizer, 6.
・Pa...Speaker, 21...AD converter, 2
2...PAR (30R analyzer, 23...
・Pitch extractor, 24... Voiced/unvoiced determiner, 2
5... Silent section detector, 26... Residual extraction circuit, 27... Amplitude determining circuit, 28...
...Encoder, 29...Switcher, 41...
・・Selector, 42・・・・）・Fumemeter memory, 4
3... Sound source memory, 44... Silent sound source generator, 46... Digital filter, 47...
...DM converter. Name of agent: Patent attorney Toshio Nakao and 1 other person68
1

Claims (2)

[Claims] (1) A speech analysis and synthesis device characterized in that a part of the residual waveform obtained by analyzing a stationary voiced sound section is transmitted as voiced driving sound source data to a silent section and synthesized into speech. (2) The speech analysis and synthesis device 0 according to claim 1, which synthesizes speech using a predetermined waveform as a voiced driving sound source data before a part of the residual waveform is extracted.