JPS5850357B2 - Speech analysis and synthesis device - Google Patents

Abstract

PURPOSE:To output the audio having naturality audibly, by measuring the short time average power of audio signal mixing surrounding noise and that for surrounding noise, subtracting the average power of surrounding noise from the average power of audio signal, and determining the amplitude of the excited power supply from the subtracted power. CONSTITUTION:The audio signal from the audio signal source 205 including the noise from the noise sound source 206 is measured to measure the parameter including frequency spectrum and audio information, and the audio is synthesized from the parameter value. In such an audio analysis and synthesis unit, the input signal of the signal source 205 including noise and the noise source 206 is fed to the short time average power measuring units 211 and 215 provided with the analyser side 202, the short time average power is measured, and the measuement value of noise is subtracted from the measurement value of audio signal with the audio power 213 and it is fed to the coder 212. The output at the analyzer side 2 with this constitution is fed to the composite coder at the synthesis side 203 and the output of the composite coder 216 is fed as the exciting sound source information of the audio synthesis filter 223.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention uses coefficients equivalent to correlation coefficients such as autocorrelation coefficients, cross-correlation coefficients, or covariance coefficients (hereinafter collectively referred to as "correlation coefficients"). A pitch-excitation type speech analysis and synthesis device that directly analyzes spectrum information of a speech signal and synthesizes a speech signal using the analysis data, such as a vocoder device such as an autocorrelation vocoder or a cross-correlation vocoder, and a correlation coefficient. Regarding a pitch excitation type voice analysis and synthesis device, for example, a linear predictive vocoder device 1, which analyzes spectrum information of a voice signal using coefficients converted from , and synthesizes a voice signal using the analysis data, there is one particular feature. This invention relates to a speech analysis and synthesis apparatus for analyzing and synthesizing speech and the like in a high ambient noise environment to obtain a suitable synthesized speech.

[Conventionally, this type of speech analysis/synthesis device] In order to transmit information of the analyzed waveform from the analysis side to the synthesis side, spectrum information analyzed using the correlation coefficient directly or correlation coefficient is used. In addition to spectrum information analyzed using coefficients converted from relational coefficients, pitch period information, voiced/unvoiced discrimination information, short-term average power information, etc. were transmitted.

As is well known, the frequency spectrum of a noisy audio signal analyzed using a correlation coefficient can be analyzed by suppressing or flattening the noise component and emphasizing the extremely thick frequency components (for example, formant frequencies) of the audio signal component. There is a tendency to

Example Eva 1 In Figure 1, an audio signal with a spectrum envelope 101 has a spectrum envelope 1
The spectrum envelope of a signal mixed with noise having a value of 02 is originally expressed as 103, but if it is expressed using a correlation coefficient, for example, with an all-pole model, 1 is specially expressed as 104. This is transformed into a shape that more closely resembles the spectrum of the voice itself.

One of the effects of the spectrum modification 1 is that, for example, when analyzing a mixed signal of a speech signal and noise with approximately equal short-term average power, the formant bandwidth expressed from the analysis result is similar to that of the speech itself. There may be cases where the band width of the formant is almost the same as that of the formant.

Normally, the amplitude of the synthesized speech in this type of speech analysis and synthesis apparatus is determined by the bandwidth of the formant, the amplitude of the excitation sound source added to the filter E for generating the synthesis, etc.

Conventionally, in this type of speech analysis and synthesis apparatus 1, when the input speech signal level is one level higher than ambient noise, for example, the amplitude reproducibility of the synthesized speech is good.

However, if the input audio signal level is about the same as the ambient noise level,
Alternatively, if the amplitude of the excitation sound source added to the filter for generating synthesized speech is below the ambient noise level (although it is strongly influenced by the ambient noise power),
Since the formant band is close to the formant band of the input audio signal itself, the amplitude of the synthesized speech increases by an abnormality and the amplitude reproducibility deteriorates significantly.

Generally speaking, the ambient noise level is limited for a short period of time (for example, several tens of meters5)
EC to several 5 EC), it does not change much, but the audio signal level often changes drastically within a short period of time.

For example, the level differs greatly between a voiced sound stationary part and a voiced sound ending.

Therefore, in this type of speech analysis and synthesis device 1, for example, a low-level four-tone sound near the end is relatively emphasized by 1, rather than a relatively large voice sound stationary part of a 1-level.
It has the disadvantage of significantly impairing the naturalness of the auditory sense.
there was.

An object of the present invention is to provide a speech analysis and synthesis device 1 that analyzes and synthesizes speech using a correlation coefficient measured from an input signal, and uses the short-term average power of ambient noise and the output of a speech signal mixed with ambient noise. measuring the short-time average power, subtracting the short-time average power of the ambient noise from the short-term average power of the audio signal mixed with the ambient noise, and determining the amplitude of the excitation sound source using the subtracted power. Therefore, it is an object of the present invention to provide a device capable of reproducing good synthesized speech by eliminating the above-mentioned drawbacks.
There it is.

Next, a first embodiment of the present invention will be described with reference to the drawings.

FIG. 2 is a book diagram for explaining the first embodiment of the present invention.

In FIG. 2, a portion surrounded by a dashed line 201 shows the structure of the present invention.

The part surrounded by a dotted line 202 shows the analysis side of the speech analysis and synthesis apparatus according to the first invention, and the part surrounded by a dotted line 203 similarly shows the synthesis side.

The sound wave signals from the audio signal source 205 and the noise source 206 are input to the acoustic signal-to-electrical signal converter A20γ.

The acoustic signal-to-electrical signal converter A20γ converts the sound wave signal into one electrical signal, and converts it into a noise-containing audio signal, which is sent to the correlation coefficient spectrum information analyzer 208 and the voiced/unvoiced discriminator 2.
09, pitch extractor 210, and short-time average power meter A2
11 and outputs the converted electrical signal.

The correlation coefficient type spectrum information analyzer 208 performs direct or indirect spectrum analysis on the noise-containing audio signal using correlation coefficients, and outputs the analysis result to the encoder 212.

The voiced/unvoiced discriminator 209 determines whether the voice signal containing noise is voiced or not, and outputs the determination result to the encoder 212 .

A pitch extractor 210 extracts a pinch period from a noise-containing audio signal using a pinch extraction method 1 such as an autocorrelation method, and outputs pitch period information to an encoder 212.

The short-time average power measuring device A212 measures the short-term average power of the audio signal mixed with noise, and outputs the measurement result to the audio power measuring device 213.

The acoustic signal from the noise source 206 is input to the acoustic signal-to-electrical signal converter B214.

The acoustic signal-to-electrical signal converter B214 converts the sonic signal into one electric signal, and outputs the converted electric signal as a noise signal to the short-time average power meter B215.

The short-time average power measuring device B215 measures the short-time average power of the noise signal and outputs the measurement result to the audio power measuring device 213.

The audio power measuring device 213 is a short-time average power measuring device A211.
The difference between the short-time average power of the noise-containing audio signal supplied from the B215 and the short-time average power of the noise signal supplied from the short-time average power meter B215 is measured, and the measurement result is used as the audio power. Output to encoder 212.

The analysis result supplied from the encoder 212G1 correlation coefficient type spectrum information analyzer 208 and the voiced/unvoiced discriminator 20
9, the pitch period information supplied from the pitch extractor 210, and the audio power supplied from the audio power measuring device 213 are each quantized and multiplexed, and the multiplexed signals are sent to the transmission path. decoder 216 via 204
Output to.

The decoder 216 separates and decodes the multiplexed signal, sends the decoded spectrum information to the spectrum information transmission line 211, and sends the decoded voiced/unvoiced discrimination information to the voiced/unvoiced discrimination information transmission line 21.
8, decoded high-notch cycle information to pitch transmission line 219,
The decoded audio power information is output to the power information transmission path 220.

The sound source signal generator 221 is connected to the voiced/unvoiced discrimination information transmission line 218
From the decoded voiced/unvoiced discrimination information supplied from the decoded voiced/unvoiced discrimination information, the decoded pitch cycle information supplied from the pitch transmission line 219, and the decoded voice power information supplied from the power information transmission line 220, a channel vocoder known as a Datsudray vocoder, for example, is used. Excitation signal generation method 1 Generate an excitation signal from
The excitation signal is output to the excitation signal transmission path 222.

The speech synthesis filter 223 is connected to the spectrum information transmission path 21
The filter coefficients are determined from the decoded spectrum information E supplied via the excitation signal transmission line 222, and the output is outputted as a synthesized audio signal to the waveform output terminal 224. Ru.

Next, a second embodiment of the present invention will be described with reference to the drawings.

FIG. 3 is a block diagram for explaining a second embodiment of the present invention.

FIG. 3 shows a part of the configuration of the analysis side in the first aspect of the present invention,
A portion surrounded by a dashed line 301 shows the configuration of the present invention.

FIG. 3 shows an embodiment in which the present invention is applied to a so-called breath-talk type voice communication device I.

When the audio signal source 302 generates audio, a transmitting signal is always input to the control signal input terminal 304.

Furthermore, when the speech interruption signal is input to the control signal input terminal 304, the audio signal source 302 does not generate audio.

When the speech interrupt signal is input to the control signal input terminal 304, the sound wave signal from the noise source 303 is input to the acoustic signal-electrical signal converter 305.

The acoustic signal-to-electrical signal converter 305 converts the sonic signal into an electrical signal and outputs it to the short-time average power measuring device 306 .

The short-time average power measuring device 306 measures the short-time average power of noise and outputs it to the memory 301.

The storage device 30γ stores the short-time average power when the speech interruption signal is inputted through the control signal input terminal 304.

When a transmission signal is input to the control signal input terminal 304, the sound wave signals from the audio signal source 302 and the noise source 303 are input to the acoustic signal-electrical signal converter 305.

The acoustic signal-to-electrical signal converter 305 converts the acoustic signal into an electrical signal and outputs it to the short-time average power meter 306.

The short-time average power measuring device 306 measures the short-term average power of the audio signal mixed with noise and outputs it to the audio power measuring device 308 .

When a transmitting signal is input through the control signal input terminal 304, the memory 307 outputs the short-term average power of the noise stored in advance to the voice power meter 308.

The audio power meter 308 calculates the short-term average power of the noise-containing audio signal supplied from the short-term average power meter 306 multiplied by the short-term average power of the noise supplied from the memory 307 by a constant a. and generate an audio power signal,
It is output to the audio power signal output terminal 309.

The audio power signal output terminal 309 corresponds to the output terminal of the audio power measuring device 213 shown in FIG. 2f in the first embodiment.

Note that the constant a is determined in consideration of the short-term fluctuation rate of the noise level.

The speech analysis and synthesis device according to the first aspect of the present invention measures the short-term average power of ambient noise and the short-term average power of an audio signal mixed with ambient noise, and measures the difference between the two types of short-term average power. By doing this, the short-term average power of the audio signal is obtained, and from this, the amplitude of the excitation sound source is determined.

Therefore, it is possible to improve the amplitude reproducibility of synthesized speech by a speech analysis and synthesis device that analyzes spectrum information of speech signals using a correlation coefficient whose analysis formant band is not affected by ambient noise. be.

[Brief explanation of the drawing]

FIG. 1 is a waveform diagram for explaining characteristics of spectrum analysis using correlation coefficients, and FIGS. 2 and 3 are block diagrams showing embodiments of the present invention. 202... Analysis side, 203... Synthesis side, 204... Transmission path, 205, 302...
...Audio signal source, 206.303...Noise source,
207,214゜305...Acoustic signal-electrical signal converter, 208...Correlation coefficient type spectrum information analyzer, 209...Voiced/unvoiced discriminator, 21
0...Pitch extractor, 211.215,306
...Short-time average power meter, 212...
- Encoder, 213, 308...Audio power measuring device, 216...Decoder, 217...
Spectrum information transmission line, 218... Voiced/unvoiced discrimination information transmission line, 219... Pitch transmission line, 2
20...Power information transmission line, 221...
Sound source signal generator, 222...excitation signal transmission line,
223...Speech synthesis filter, 30γ...
...Memory device.

Claims (1)

[Claims] 1. A synthesis filter 1 that measures parameters indicating frequency spectrum and sound source information measured from an input signal consisting of an audio signal including ambient noise, and determines various coefficients and excitation input from the parameter values 1. In the speech analysis and synthesis device 1 for synthesizing speech, a first measuring means for measuring the short-term average power of the input signal, a second measuring means for measuring the short-term average power of the ambient noise component, and a third measuring means for outputting a difference between the output of the first measuring means and the output of the second measuring means, and the output of the third measuring means is used as excitation sound source information of the synthesis filter. Features of speech analysis and synthesis equipment. 2. In the speech analysis and synthesis device 1 according to claim 1, the first measuring means and the second measuring means measure the input signal and the ambient noise component from an external control signal 1. The third measuring means measures the short-term average power of the input signal obtained by the measuring means and the ambient noise component. A speech analysis and synthesis device characterized in that it is a means for measuring a difference from a short-time average power.