JPH01267700A - Speech processor - Google Patents

Abstract

PURPOSE:To obtain a smooth speech which contains no noise by editing the output of a reverse filter in each pitch cycle which is the output of a pitch extractor and leading the output to a synthesizing filter, and determining the characteristics of the reverse filter and synthesizing filter with the output of a spectrum analyzer. CONSTITUTION:The output of the reverse filter 8 is edited in each pitch cycle which is the output of the pitch extractor 6 and led to the synthesizing filter 10. The characteristics of the reverse filter 8 and composite filter 10 are determined with the output of the spectrum analyzer 7. Then the output of an editing device 9 is inputted to the synthesizing filter 10 (having the opposite characteristics from the reverse filter 8 because of the output of the spectrum analyzer 7) to obtain a speech signal wherein phoneme information is restored. Their input/output speed ratios need not be integers or reciprocals like 2 and 1/3 and may be 1.2 and 1.1, and the object of editing is a residue signal having no correlation, so there is no noise. Consequently, the smooth speech having no noise can be obtained.

Description

[Detailed Description of the Invention] [Industrial Field of Application] The present invention allows rapid listening without changing the utterance phoneme even if the tape speed is changed during playback by a tape recorder.

This invention relates to a speech processing device that allows slow listening.

[Conventional technology]

Conventionally, an example of this type of audio processing device is vSC (Va
There was a method called riable 5peach Control) that prevented changes in phonology by changing the speed at which speech waveforms were written into memory and the speed at which they were interpreted.

[λ-ma problem that the invention seeks to solve]

The vSC in the conventional audio processing device described above edits the pitch of the audio waveform, so noise is generated at the joints of the waveforms, which is very noticeable. There was a division.

[Means to solve the problem]

The audio processing device of the present invention has two time base clock signal input terminals, a first clock signal and a second clock signal, and includes a pitch extractor, a spectrum analyzer, and an inverse filter each receiving an audio signal as input. The synthesis filter is configured to operate with the first clock signal, the synthesis filter is configured to operate with the second clock signal, and the output of the inverse filter is collected every Km of pitch synchronization, which is the output of the pitch extractor. The inverse filter is guided to the synthesis filter, and the power of the spectrum analyzer is configured to determine the characteristics of the inverse filter and the synthesis filter.

[Effect]

In the present invention, smooth speech without noise is obtained by editing according to the pitch period at the residual stage.

〔Example〕

Hereinafter, embodiments of the invention will be explained in detail with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention.

In the figure, 1 is an audio input terminal. 2 is a speed signal input terminal, 3 is a read speed signal input/terminal, and these are two of the first clock signal and second clock signal.
It constitutes two time base clock signal input terminals. 4 is the audio output terminal.

5 is an A/D converter, and 6, 7.8 are a pitch extractor, a spectrum analyzer, and an inverse filter that receive the audio signal output from the A/D converter 5, respectively. 7 and the inverse filter 8 are configured to operate with a clock signal from the speed signal input terminal 2. 9 is an editor; 10 is a synthesis filter; this synthesis filter 10 is configured to operate with a clock signal from the read speed signal input terminal 3; 11 is a D/A converter.

The output of the inverse filter 8 is edited for each pitch period, which is the output of the pitch extractor 6, and is guided to the synthesis filter 10, and the output of the spectrum analyzer 7 determines the characteristics of the inverse filter 8 and the synthesis filter 10. It is structured in such a way that it makes a decision.

Figures 2 and 3 are diagrams that provide the operation theory BAK of Figure 1. Figure 2 shows the case where the edited waveform manual speed is greater than the readout speed, and Figure 3 shows the edited waveform. This shows the case of input speed <read speed. Note that (a) indicates the input residual, and (a) indicates the output residual.

Next, the operation of the embodiment shown in the first rlAK will be explained with reference to FIGS. 2 and 3.

First, an audio signal applied to the audio input terminal 1 is converted into a digital signal by the A/D converter 5.

The rate to be converted is determined by a clock signal proportional to the tape speed of the tape recorder input to the speed signal input terminal 2. The digitized speech is then guided to a pitch extractor 6, a spectrum analyzer 7, and an inverse filter 8, respectively. Here, the pitch extractor 6, spectrum analyzer T and inverse filter 8 are connected to the A/D converter 5.
It operates with the same timing clock.

Next, a residual signal is obtained by an inverse filter 8 using the spectrum signal obtained by the spectrum analyzer T.

This residual signal is led to an editor 9, and edited by the pip extractor 6 in pitch period units.

Now, a clock signal of nKH2 is input from the speed signal input terminal 2, and a clock signal of mKH2 is input from the read speed signal input terminal.
When a Hz clock signal is input and n>m, one pitch period of the residual signal input to the editor 9 is cut out, expanded by n7m times, recombined, and output. Ru. This case of n=2m is shown in g2[ffi. In this Figure 2, (a) indicates the part to be discarded.

Then, the output of the editor 9 is inputted to a synthesis filter 10 (which has characteristics opposite to those of the inverse filter 8 based on the output of the spectrum analyzer 7), and a speech signal with phoneme information restored is obtained. When n>m, fast speech is output without raising the pitch.

In the case of reverse Kn and <rn, @W is performed when the tape speed is slower than the standard. In this case, as shown in FIG. 3, the input residual (a) is reduced by n7m times, reproduced, and guided to the synthesis filter 10.

Figures 2 and 3 show cases where the human power/output speed ratio is 2 and 1/3, respectively, but the ratio need not be an integer or its reciprocal; It doesn't matter if it is 1.1.

And since the object of editing is the uncorrelated residual signal,
There's no noise.

For convenience of explanation, an all-digital example using an A/D converter and a D/A converter is shown.
The values do not necessarily have to be digital (it is also easy to create a system with discrete time readings using an analog recording device such as a BBD).

In addition, the speed signal input terminal 2 is linked to the tape speed, and the read speed signal input terminal 3 is fixed, which is the normal usage for a tape recorder that can be heard even when the tape is sped up or slowed down. By making the read speed signal input terminal 3 variable while keeping the input terminal 2 fixed, it is possible to create a voice with a different timbre and to create special effects.

〔Effect of the invention〕

As explained above, the present invention has the advantage that smooth speech without noise can be obtained by editing according to the pip period at the residual stage. You can also create unusual voices and create special effects.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention, and FIGS. 2 and 3 are diagrams for explaining the operation of FIG. 1. 1...Audio input terminal, 2...Speed signal input terminal, 3...Reading speed signal input terminal, 6...
・Pitch extractor, T...Speech analyzer, 8...
... Inverse filter, 9... Editor, 10... Synthesis filter. Patent applicant: NEC Corporation

Claims (1)

[Claims] A pitch extractor, a spectrum analyzer, and an inverse filter each having two time base clock signal input terminals, a first clock signal and a second clock signal, each receiving an audio signal, operate on the first clock signal. The synthesis filter is configured to operate with the second clock signal, and the output of the inverse filter is edited for each pitch period that is the output of the pitch extractor and guided to the synthesis filter. and an output of the spectrum analyzer determines characteristics of the inverse filter and the synthesis filter.