JPS59187400A - Voice analyzer/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] The present invention relates to a speech analysis and synthesis device.

Analyzes a digitized audio waveform (hereinafter referred to as a digitized waveform) to express parameter values representing its spectral envelope (hereinafter referred to as spectral information) and characteristics of the sound source such as pitch, voiced/unvoiced, amplitude, etc. While extracting parameter values (hereinafter referred to as sound source information) and controlling a digital synthesis filter (hereinafter referred to as synthesis filter) that forms a spectral envelope based on the spectral information, the sound source waveform generated based on the sound source information is controlled. 2. Description of the Related Art Among speech analysis and synthesis apparatuses that synthesize speech by inputting it to the synthesis filter, there is known one that utilizes a residual waveform as a sound source waveform in order to improve the sound quality of the synthesized speech. That is, at the time of analysis, the audio waveform is input to an analysis filter that is an inverse filter to the synthesis filter,
From the output waveform (hereinafter referred to as the residual waveform), a representative residual segment waveform with a time length approximately below the -bit cycle is extracted and used as the sound source waveform that is input to the synthesis filter during synthesis. It is. In this conventional example, for voiced sounds, a synthesized sound with relatively good clarity can be obtained by repeatedly inputting the representative residual segment waveform to the synthesis filter every pitch period, but for unvoiced sounds, a synthesized sound with relatively good clarity can be obtained. It had the disadvantage of not being suitable.

In order to improve the sound quality by using the residual waveform even for unvoiced sounds, in the case of unvoiced sounds, the sample values of the representative residual segmental waveform are randomly rearranged every repetition and input to the synthesis filter. Devices that synthesize speech are known. However, in this second conventional example, the control for randomly rearranging the sample values of the representative residual segment waveform is complicated υ, and the spectrum of the waveform obtained by randomly rearranging the sample values of the representative residual segment waveform is still complicated. However, there is a drawback in that the spectrum is not necessarily related to the spectrum of the original representative residual elemental waveform, and therefore the sound quality of the synthesized sound is not necessarily improved.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech analysis/synthesis device that has a relatively simple synthesis section and that can obtain high-quality synthesized speech by effectively utilizing the information of the residual waveform regardless of whether it is voiced or unvoiced. . That is, the present invention provides a means for extracting a spectral envelope parameter value of an input voice, a means for extracting voiced/unvoiced, pinch, or amplitude of a sound source, an analysis filter controlled based on the spectral envelope parameter value, and a means for extracting a spectral envelope parameter value of an input voice. An analysis unit is configured from a means for extracting a representative residual segment waveform from a residual waveform that is an output when inputting,
means for generating an impulse train sound source waveform of the pitch extracted in the synthesis section; means for generating a pseudo white noise sound source waveform; means for switching the noise source waveform; means for adjusting the amplitude of the sound source waveform based on the amplitude information extracted in the analysis section; a transversal filter that filters the sound source waveform, and a transfer function that is controlled based on the spectral envelope parameter value extracted in the analysis section, and the output waveform of the transversal filter is The main feature is that the synthesis section is composed of a synthesis filter for filtering.

Next, one embodiment of the present invention will be described using the drawings.

The figure is a block diagram of one embodiment of the present invention.

In the analysis section 1, when an audio waveform is input from the input terminal 8, the spectral information extraction circuit 5 first extracts the spectral information of the audio waveform according to the control information sent from the analysis section control circuit 3 via the control information transmission line 24. It is extracted and outputted to the spectrum information output terminal 10 and sent to the analysis filter 4 via the spectrum information transmission line 27. This spectral information extraction circuit 3 can be realized by a linear prediction method or a formant analysis method.

On the other hand, the sound source information extraction circuit 6 extracts the pitch, voiced/unvoiced, and amplitude values of the audio waveform according to the control information sent from the analysis section control circuit 3 via the control information transmission line 23 and sends the extracted values to the sound source information output terminal 11. Output.

The analysis filter 4 filters the input audio waveform based on the spectrum information sent from the spectrum information extraction circuit 5 in accordance with the control information sent from the analysis section control circuit 3 via the control information transmission path 25, and outputs the output. It is sent to the representative residual segment waveform extraction circuit 7 via the residual waveform transmission path 28. Note that this analysis filter has an inverse filter relationship with the synthesis filter 18, which will be described later.

The representative residual segment waveform extraction circuit 7 extracts a predetermined number of representative samples from the residual waveform sent from the analysis filter 4 in accordance with control information sent from the analysis unit control circuit 3 via the control information transmission path 26. The residual segment waveform is extracted and output to the representative residual segment waveform output terminal 9.

The analysis unit control circuit 3 instructs the analysis filter 4 and the extraction circuits 5, 6.7 of the third heel to analyze and extract parameter values at certain time intervals, for example, from several milliseconds to about 20 milliseconds. Instruct.

The audio waveform is analyzed and extracted in the analysis section 1, and each piece of information outputted from the three output terminals 9, 10, and 11 is temporarily stored or sent via a transmission path to the synthesis section 203. The signal is input to the small 0 input terminal, that is, the representative residual segment waveform input terminal 19, the spectrum information input terminal 20, and the sound source information input terminal 21.

In the synthesis section 2, the three input terminals 19, 20.2
The information input from 1 is sent to a transversal filter 17, a synthesis filter 18, and a switch 15, respectively.

The synthesis unit control circuit 12 is an impulse generation circuit 13 which will be described later.
, the white noise generation circuit 14, the sound source adjustment circuit 16, the transversal filter 17, and the synthesis filter 18 are being controlled respectively [11] Information transmission line 3LiH29Hjl+
33 + 34 for each sampling period, and instructs each circuit to generate, adjust, or filter a waveform. 111 to switch the sound source information via the control information transmission path 32.
The impulse generation circuit 13 is instructed to change the pitch parameter value fJ-i, and the sound source adjustment circuit 16 is instructed to change the voiced/unvoiced and amplitude parameter values, and the transformer is still The versatile filter 17 and the synthesis filter 18 are instructed to update the coefficients of the transfer function.

The switch 15 transmits the pitch parameter value of the sound source information input from the sound source information input terminal 21 to the sound source information transmission line 35 in accordance with the control information sent from the synthesis section control circuit 12.
to the impulse generating circuit 13 via the voiced and unvoiced
and amplitude parameter values are sent to the sound source adjustment circuit 16 via the sound source information transmission line 36.

The impulse generation circuit 13 generates an impulse train having a period equal to the pitch parameter value sent via the switch 15 in accordance with the control information sent from the synthesis section control circuit 12, and adjusts the sound source via the impulse sound source transmission line 37. to circuit 16.

On the other hand, the white noise generation circuit 14 generates pseudo white noise according to the control information sent from the synthesis section control circuit 14, and sends it to the sound source adjustment circuit 16 via the white noise sound source transmission line 38.

The sound source adjustment circuit 16 adjusts the waveform sent from the impulse generation circuit 13 or the white noise generation circuit 14 according to the control information sent from the synthesis section control circuit 12 and the voiced/unvoiced parameter value sent via the switch 15. is selected, the amplitude is adjusted according to the amplitude parameter value sent via the switch 15, and is sent to the transnomous filter 17 via the sound source waveform transmission path 39.

The transversal filter 17 is connected to the synthesis section control circuit 1.
2, the sample value of the representative residual segment waveform input from the representative residual segment waveform input terminal 19 is used as a coefficient of a transfer function, and the sound source waveform sent from the sound source adjustment circuit 16 is filtered. and sends it to the synthesis filter 18 via the sound source waveform transmission line 40.

The synthesis filter 18 determines the coefficients of the transfer function according to the spectral information input from the spectral information input terminal 20 in accordance with the control information sent from the synthesis unit control circuit 12, Filter the waveform sent from 17 and output the result to terminal 2
Output to 2.

As explained above, according to the present invention, the control regarding voiced and unvoiced is simply configured by switching between two types of sound source waveforms. This has the effect that a sound source waveform having a spectral shape can be reproduced and a high-quality synthesized sound can be obtained.

[Brief explanation of drawings]

The figure is a block diagram of one embodiment of the present invention. 1... Analysis section 2... Synthesis section 3...
Analysis section control circuit 4...Analysis filter 5...χ
W) Tit information extraction circuit 6... Sound source information extraction circuit 7... Representative residual segment waveform extraction circuit 8... Input terminal 9-... Representative residual segment waveform output terminal 10...
Spectrum information output terminal 11...Sound source information output terminal
12... Synthesis unit control circuit 13... Impulse generation circuit 14... White noise generation circuit 15... Switch 16... Sound source adjustment circuit 17... Transversal filter 18...-Synthesis filter 19
... Representative residual segmental waveform input terminal 20 ... Spectrum information input terminal 21 ... Sound source information input terminal 22
--Output terminal patent applicant NEC Corporation

Claims (1)

[Claims] (1) Analyze the digitized audio waveform to extract spectrum information representing its spectral envelope and sound source information representing characteristics of the sound source such as pitch, voiced/unvoiced, amplitude, etc. A speech analysis and synthesis device that generates a sound source waveform and inputs it to a digital synthesis filter controlled based on the spectral information to synthesize speech, which is controlled based on the spectral envelope information and filters the input speech waveform. The analysis section includes a digital analysis filter and a means for extracting a representative residual segment waveform from the residual waveform obtained as an output of the filter, and the sample value of the representative residual segment waveform is used as a coefficient of a transfer function. Equipped with a digital transversal filter in the synthesis section,
A speech analysis and synthesis device, characterized in that the sound source waveform is filtered by the digital transversal filter, and the output thereof is input to the digital synthesis filter.