JPH03236000A - Audio signal processor - Google Patents

Abstract

PURPOSE:To remove noise and to improve articulation by detecting the vowel and consonant areas of an audio signal where the noise is mixed, setting a cancel coefficient according to the detection result, and removing a predicted noise component. CONSTITUTION:A cepstrum analysis of the frequency analytic output of the audio input signal is taken to detect the cepstrum peak and also calculate the mean value level, a vowel/consonant decision means 5 decides the vowels and consonants according to peak detection information and mean value information, and a cancel coefficient setting means 17 utilizes the decision results to set the cancel coefficient. A noise predicting means 6, on the other hand, predict noise components of the band-divided audio signal, a canceling means 8 inputs the noise prediction output, audio signal, and cancel coefficient signal to cancel the noise components in consideration of the cancel rate, and a signal composing means 9 composes its cancel output. Consequently, the noise is suppressed and the audio signal processor with good articulation is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to an audio signal processing device that can be used for audio processing and the like.

Prior Art FIG. 8 is a block diagram of a conventional signal processing device. 1
1 is a filter control unit to which a signal mixed with noise is input and detects the signal or noise; 12 is a BPF group having a large number of bandpass filters; and 13 is an adder. That is, the filter control unit 11 controls the filter coefficients of the BPF group 12 according to the noise or signal of the input signal, and
The PF group 12 is a group of band-pass filters, and is configured to divide an input signal into appropriate bands, and to determine the pass band characteristics according to a control signal from the filter control section 11.

The operation of the conventional signal processing device configured as described above will be described below.

The input signal with noise superimposed on the voice is sent to the filter control unit 11.
is supplied to The filter control unit 11 obtains noise components from the supplied signal corresponding to each band of the BPF group 12, and
Filter coefficients that do not allow noise components to pass through the PF group 12 are supplied to the BPF group 12.

The BPF group 12 divides the input signal into appropriate bands, passes the input signal appropriately according to the filter coefficients inputted from the filter control unit 1 for each band, and supplies the input signal to the adder 13. The adder 13 mixes the signals divided into appropriate bands by the BPF group 12 to obtain an output.

By doing this, the BPF group 12 lowers the pass level of the input signal in a band containing noise. As a result, a signal with attenuated noise components is obtained.

Problems to be Solved by the Invention However, the amount of noise and the clarity do not necessarily match, and therefore, in conventional signal processing devices, although noise can be suppressed, there is a problem that the clarity does not improve.

The present invention solves the problems of conventional signal processing devices, and aims to provide a speech signal processing device that suppresses noise and provides good clarity while determining vowels/consonants. .

Means for Solving the Problems The present invention as claimed in claim 1 provides a frequency analysis means for frequency analyzing an audio input signal, a cepstrum analysis means for cepstrum analysis of the frequency analysis output of the frequency analysis means, and a cepstrum analysis means for cepstrum analysis of the frequency analysis output of the frequency analysis means. a peak detection means for detecting a kebnutrum peak in the analysis output; an average value calculation means for calculating an average level in the cepstrum analysis output of the cepstrum analysis means; and peak detection information of the peak detection means and the average value calculation means. Determine the vowel based on the peak based on the average value information of
Vowel/consonant determining means for determining whether a consonant is a vowel or a consonant by determining a consonant based on the level of the average value information, and a cancellation coefficient setting means for setting a cancellation coefficient using the determination result of the vowel/consonant determining means. and a noise prediction means which receives the Fourier-transformed audio signal and predicts its noise component, a noise prediction output of the noise prediction means, the audio signal, and a cancellation coefficient signal set by the cancellation coefficient setting means. is input, and the signal processing device is characterized by comprising a canceling means for canceling a noise component in consideration of the cancellation rate from the audio signal, and a signal synthesizing means for synthesizing the canceling output of the canceling means. .

The present invention as set forth in claim 2 provides a band division means for Fourier transforming an audio input signal, a cepstrum analysis means for cepstrum analysis of the band division output of the band division means, and a cepstrum peak in the cepstrum analysis output of the cepstrum analysis means. Based on the peak detection means to detect, the average value calculation means to calculate the average level in the cepstrum analysis output of the cepstrum analysis means, the peak detection information of the peak detection means and the average value information of the average value calculation means, determining a vowel based on the peak;
Vowel/consonant determining means for determining whether a consonant is a vowel or a consonant by determining a consonant based on the level of the average value information, and a cancellation coefficient setting means for setting a cancellation coefficient using the determination result of the vowel/consonant determining means. and a noise prediction means which receives the Fourier-transformed audio signal and predicts its noise component, a noise prediction output of the noise prediction means, the audio signal, and a cancellation coefficient signal set by the cancellation coefficient setting means. is input, and is equipped with a canceling means for canceling a noise component from the audio signal in consideration of its cancellation rate, and a band synthesizing means for band synthesizing the cancellation output of the canceling means. be.

Claim 3 The present invention provides the vowel/
The consonant determining means includes a first comparator that compares at least the peak detected by the peak detecting means and the Schlesbold value set by the Schleshold setting section, and the average value and Schleshold value calculated by the mean value calculating means. A second comparator that compares a predetermined Schlesbold value set in the setting section, and based on the comparison results of the first and second comparators,
This is an audio signal processing device characterized by comprising a vowel/consonant determination circuit that determines vowels and consonants and outputs the results.

In the present invention, the frequency analysis means frequency-analyzes the audio input signal, the cepstrum analysis means cepstrum-analyzes the frequency analysis output of the frequency analysis means, and the peak detection means detects the cepstrum peak in the cepstrum analysis output of the cepstrum analysis means. and the average value calculation means calculates the average level in the cepstrum analysis output of the cepstrum analysis means, and the vowel/consonant determination means is based on the peak detection information of the peak detection means and the average value information of the average value calculation means. determining a vowel based on the peak, determining a consonant based on the level of the average value information,
Vowels and consonants are determined, and the cancellation coefficient setting means determines vowels/consonants.
A cancellation coefficient is set using the determination result of the consonant determination means. On the other hand, the noise prediction means is inputted with the band-divided audio signal and predicts its noise component, and the cancellation means is set based on the noise prediction output of the noise prediction means, the audio signal, and the cancellation coefficient setting means. A cancellation coefficient signal is input, noise components are canceled from the audio signal in consideration of the cancellation rate, and the signal synthesis means synthesizes the cancellation output of the cancellation means.

Example Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an audio signal processing device in one embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a band division means for dividing a signal into frequency bands, in particular an FFT means for Fourier transforming the signal, as an example of a frequency analysis means for performing frequency analysis on a signal, and 2 a cepstrum for performing cepstral analysis. Analysis means; 3 is a peak detection means for detecting the peak of the cepstrum distribution; 4 is an average value calculation means for the cepstrum distribution; 5 is a vowel/consonant determination means for determining vowels and consonants. That is, the FFT means 1 performs fast Fourier transform on the audio signal input and supplies it to the cepstrum lO- analysis means 2. The cepstrum analysis means 2 obtains the cepstrum of the spectrum signal and supplies it to the peak detection means 3 and the average value calculation means 4. Second
This is shown in Figures (a) and (b). The peak detection means 3 determines the peak of the cepstrum obtained by the cepstrum analysis means 2 and supplies it to the vowel/consonant determination means 5.

On the other hand, the average value calculation means 4 calculates the average value of the cepstrum obtained by the cepstrum analysis means 2, and supplies it to the vowel/consonant determination means 5. The vowel/consonant determination means 5 determines vowels and consonants of the audio signal input using the peak of the cepstrum supplied from the peak detection means 3 and the average value of the cepstrum supplied from the average value calculation means 4, and calculates the determination result. This is used as a judgment output. 6 is a noise prediction means for inputting the output signal of the FFT 1 and predicting a noise component; 8 is a canceling means for removing noise as described later; 9
is an example of a signal synthesizing means, which is below the band synthesizing means, especially IFFFTFF which performs inverse Fourier transform.

To explain in detail, the noise prediction means 2 is a means for predicting noise components for each channel based on the voice/noise input divided into m channels 11-channels, and supplying the predicted noise components to the cancellation means 3. For example, the noise prediction is as shown in FIG. That is, the y-axis shows the frequency, the y-axis shows the audio level, and the z-axis shows the time, and the data p1, p2, etc. at the frequency f1 are plotted. ..., take pl and predict pJ beyond that.

For example, the average of the noise parts pl-pi is taken as pJ.

Alternatively, when the audio signal portion continues, pJ is multiplied by an attenuation coefficient. Further, the canceling means 8 is
The m-channel signals are supplied from the FT 1 and the noise prediction means 6, and are canceled by subtracting noise for each channel according to the input of the cancellation coefficient, and then sent to the IFFT means 9.
It is a means of supplying That is, the predicted noise component is multiplied by a cancellation coefficient to cancel it. Generally, as an example of a cancellation method, cancellation on the time axis is a method of subtracting a predicted noise waveform (b) from a noise-containing audio signal (a), as shown in FIG. As a result, only the signal is extracted (c).

12- Also, as shown in Fig. 5, cancellation based on the frequency is performed by Fourier transforming the noisy speech signal (a), +-ff), and then calculating the predicted noise spectrum (
(c) is subtracted (d) and subjected to inverse Fourier transform to obtain a noise-free audio signal (e). IFFT means 9
is a means for obtaining an audio output by performing inverse Fourier transform on the m-channel signals supplied from the canceling means 8.

The cancellation coefficient setting means 7 is a means for appropriately setting a cancellation coefficient using the vowel/consonant area information determined by the vowel/consonant determination means 5. For example, in the voice area, in order to ensure clarity, the cancellation coefficient is set to be small in order to improve intelligibility by not intentionally removing noise components, and in other noise areas, the cancellation coefficient is set to a small value, and in other noise areas, noise components are completely eliminated. In order to eliminate this, the cancellation coefficient may be increased. In the present invention, not only vowels but also consonants are reliably detected, so the clarity of the obtained speech is sufficiently good.

The operation of the audio signal processing device according to the embodiment of the present invention configured as described above will be described below.

The audio signal input is fast Fourier transformed by FFT means 1,
Next, the cepstrum is determined by the cepstrum analysis means 2, and the peak of the cepstrum is determined by the peak detection means 3. Further, the average value of the cepstrum is calculated by the average value calculating means 4. When the vowel/consonant determining means 5 receives a signal indicating that a peak has been detected from the peak detecting means 3, it determines that the audio signal input is in a vowel section. Regarding the determination of consonants, for example, if the cepstrum average value inputted from the average value calculating means 4 is larger than a predetermined value, or if the amount of increase (differential coefficient) of the cepstrum average value is larger than a predetermined value, If it is larger than the specified value, it is determined that the audio signal input is in a consonant section. As a result, a signal indicating a vowel/consonant or a signal indicating a speech section including a vowel and a consonant is output.

On the other hand, the voice/noise input containing noise is processed by the noise prediction means 6.
Then, the noise component is predicted for each channel. Further, the noise component supplied from the noise prediction means 2 for each channel is removed from the voice/noise signal by the cancellation means 3. The noise removal rate at this time is appropriately set by inputting a cancellation coefficient to improve clarity for each channel.

For example, as mentioned above, in order to ensure intelligibility in the voice area, the cancellation coefficient is set to a small value in order to improve intelligibility by not intentionally removing noise components, and in other noise areas, In order to completely remove noise components, the cancellation coefficient is set to a large value. In the present invention, not only vowels but also consonants are reliably detected, so the clarity of the obtained speech is sufficiently good.

Then, the IFFT means 9 performs inverse Fourier transform on the m-channel signal from which noise has been removed obtained by the canceling means 8 and outputs it as an audio signal.

As described above, according to this embodiment, the noise removal rate of the canceling means 8 can be appropriately given for each band by inputting a cancellation coefficient, and by selecting the cancellation coefficient with high accuracy in correspondence with the voice, - Obtain clear, noise-suppressed audio output.

Next, another aspect of the present invention will be explained.

FIG. 6 is a block diagram showing one embodiment thereof. 1st
Means that are the same as those in the illustrated embodiment are given the same numbers. That is, 1 is F for fast Fourier transform of the audio signal.
F7 means; 2 is a cepstrum analysis means for determining the cepstrum of the Fourier-transformed spectrum signal; 3 is a peak detection means for determining the peak based on the cepstrum analysis result; 4 is an average for calculating the average value of the cepstrum. 7 is a value calculation means, 6 is a noise prediction means, 7 is a cancellation means, 9 is an IFF7 means, and 7 is a cancellation coefficient setting means. The vowel/consonant determining means 5 has the following means. That is, the first comparator 52 is a circuit that compares the peak information obtained by the peak detection means 3 with a predetermined darkness value set by the first threshold setting section 51 and outputs the result. be. Also, the first
The threshold setting section 51 includes the average value calculation means 4
means 16- for setting a darkness value according to the average value obtained in the above.

Further, the second comparator 53 uses a predetermined threshold set by the second threshold setting section 54 and the average value calculation means 4.
This is a circuit that compares the average value obtained in , and outputs the result.

Further, the vowel/consonant determination circuit 55 determines whether the input audio signal is a vowel or a consonant based on the comparison result obtained by the first comparator 52 and the comparison result obtained by the second comparator 53. This is a circuit that determines whether

Next, the operation of the above embodiment will be explained.

The FFT means 1 performs fast Fourier transform on the audio signal. The cepstrum analysis means 2 obtains the cepstrum of the Fourier-transformed signal.

The peak detection means 3 detects a peak in the determined cepstrum. On the other hand, the average value calculation means 4
The average value of the cepstrum obtained above is calculated.

Next, the first threshold setting means 51 sets a threshold value as a reference for determining whether the peak obtained by the peak detection means 3 is a peak sufficient to be determined to be a vowel. At that time, the same value is determined by referring to the average value obtained by the average value calculating means 4. For example, if the average value is large, the threshold is set high to ensure that a peak indicating a vowel can be selected.

The first comparator 52 has threshold setting means 51
The set threshold value is compared with the peak detected by the peak detection means 3, and the comparison result is output.

On the other hand, the second threshold setting means 54 sets a predetermined threshold value. This may be a threshold value of the average value itself, or a dark value of a differential coefficient that indicates an increasing tendency of the average value. And the second
The comparator 53 compares the average value obtained by the average value calculating means 4 with
A comparison is made with a threshold value set by the second threshold setting means 54, and the result is output. That is, the calculated average value and the threshold average value are compared, or the increased value of the calculated average value and the threshold differential coefficient value are compared.

The vowel/consonant determination circuit 55 determines -17=18- vowels and consonants based on the comparison result of the first comparator 52 and the comparison result of the second comparator 53. If a peak is definitely detected in the comparison result of the first comparator 52, that area is determined to be a vowel. Further, in the comparison result of the second comparator 53, if the average value exceeds the average value of the threshold value, that area is determined to be a consonant. Alternatively, the increase in the average value is compared with the differential coefficient of RRlFB, and if it exceeds w7JW, it is determined to be a consonant.

The determination method of the vowel/consonant determining means 5 takes into account the nature of the vowel and consonant sections of speech, for example, the property that speech is composed of consonants + vowels, and only when the consonant section and vowel section are aligned can the first The determination output as a consonant may be output retroactively to the consonant section. That is, in order to more reliably distinguish between noise and consonants, even if a consonant is determined based on the average value, it is determined to be noise if there is no subsequent interval.

The cancellation coefficient setting means 7 sets an appropriate cancellation coefficient based on the voice information of the vowel/consonant area determined by the vowel/consonant determination means 5.

On the other hand, for voice/noise input containing noise, the noise component is predicted for each channel by the noise detector 19-measuring means 6. Further, the noise component supplied from the noise prediction means 6 for each channel is removed from the audio signal by the cancellation means 8. The noise removal rate at this time is determined by the cancellation coefficient setting means 7.
It is set for each channel by the cancellation coefficient supplied by

That is, the predicted noise component is al, and the noise mixed signal is bl.
, the cancellation coefficient is α, the output cl of the canceling means 3 becomes (bl-α, Xa:).

The cancellation coefficient α1 is a coefficient value as shown in FIG. That is, FIG. 7(a) shows the cancellation coefficient in each band. Here, fe fa indicates the entire voice/noise input band, and this fll-fa
Divide into m channels and set the cancellation coefficient. f+-f2 particularly indicates a band in which speech is included, and is reliably determined by the vowel/consonant determining means 5 as described above. In this way, in the voice band, the cancellation coefficient is made small and close to 0), and noise is not removed as much as possible. This improves clarity. human hearing 2
0- This is because the senses can hear the voice even if there is some noise.

In the non-voice bands of flI-f+ and f2fa, the cancellation coefficient is set to 1 to sufficiently remove noise. The cancellation coefficient shown in FIG. 6(b) is used when it is certain that there is no voice at all and that it can only be considered as noise, and is set to 1 to sufficiently remove noise.

For example, if a vowel does not appear at all in terms of peak frequency, it cannot be determined that it is a voice signal, so it is determined to be noise.

It is desirable to be able to switch the cancellation coefficients in FIGS. 7(a) and 7(b) as appropriate.

Note that although the present invention can be implemented in software using a computer, it can also be implemented using a dedicated hardware circuit.

As is clear from the detailed description of the invention, the audio signal processing device according to the present invention detects the vowel/consonant area of the audio signal mixed with noise, and uses the cancellation coefficient setting means based on the detected vowel/consonant area. sets the appropriate cancellation factor,
Since the predicted noise component is appropriately removed using the cancellation coefficient, the noise can be removed and the clarity can also be improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the audio signal processing device according to the present invention, FIG. 2 is a graph showing cepstrum peaks in the embodiment, and FIG. 3 explains the noise prediction method of the embodiment. 4 and 5 are waveform diagrams for explaining the cancellation method of the same embodiment,
FIG. 6 is a block diagram showing another embodiment of the audio signal processing device according to the present invention, FIG. 7 is a graph explaining the cancellation coefficient of the same embodiment, and FIG. 8 is a diagram showing a conventional audio signal processing device. FIG. 2 is a block diagram showing a processing device. 1... Frequency analysis means (band division means, FFT means)
, 2... Cepstrum analysis means, 3... Peak detection means, 4... Average value calculation means, 5... Vowel/consonant determination means, 51... First threshold setting section, 52
...First comparator, 53...Second comparator, 54...2
1-22- Second threshold setting section, 55... Vowel/consonant determination circuit, 6... Noise prediction means, 7... Cancellation coefficient setting means, 8... Cancellation means, 9... - Signal synthesis (band synthesis means, IFF'r means).

Claims (3)

[Claims] (1) A frequency analysis means for frequency analyzing an audio input signal, a cepstrum analysis means for cepstrally analyzing the frequency analysis output of the frequency analysis means, and a peak detection means for detecting a cepstrum peak in the cepstrum analysis output of the cepstrum analysis means. , an average value calculation means for calculating an average level in the cepstrum analysis output of the cepstrum analysis means, and a vowel based on the peak based on the peak detection information of the peak detection means and the average value information of the average value calculation means. a vowel/consonant determining means that determines a vowel or a consonant by determining a consonant based on the level of the average value information; and a cancellation that sets a cancellation coefficient using the determination result of the vowel/consonant determining means. a coefficient setting means, a noise prediction means which receives the Fourier transformed audio signal and predicts its noise component, a noise prediction output of the noise prediction means, the audio signal, and a noise prediction means set by the cancellation coefficient setting means. Signal processing characterized in that a cancellation coefficient signal is input, and the signal processing comprises a canceling means for canceling a noise component from the audio signal in consideration of the cancellation rate, and a signal synthesizing means for synthesizing the cancellation output of the canceling means. Device. (2) band dividing means for dividing an audio input signal into bands, cepstrum analysis means for cepstrum analysis of the band division output of the band dividing means, and peak detection means for detecting a cepstrum peak in the cepstrum analysis output of the cepstrum analysis means; , an average value calculation means for calculating an average level in the cepstrum analysis output of the cepstrum analysis means, and a vowel based on the peak based on the peak detection information of the peak detection means and the average value information of the average value calculation means. and determine the consonant based on the level of the average value information to determine the vowel/consonant.
a consonant determination means; a cancellation coefficient setting means for setting a cancellation coefficient using the determination result of the vowel/consonant determination means; and a noise prediction means for receiving the Fourier-transformed audio signal and predicting its noise component. , the noise prediction output of the noise prediction means, the audio signal, and the cancellation coefficient signal set by the cancellation coefficient setting means are input, and canceling means cancels the noise component from the audio signal in consideration of the cancellation rate. and band synthesis means for band synthesising the cancellation outputs of the cancellation means. (3) The vowel/consonant determination means includes a first comparator that compares at least the peak detected by the peak detection means and the threshold value set by the threshold setting unit;
a second comparator that compares the average value calculated by the average value calculation means and a predetermined threshold value set by the threshold value setting section; and based on the comparison results of the first and second comparators, vowels, Vowels that determine consonants and output the results/
3. The audio signal processing device according to claim 2, further comprising a consonant determination circuit.