JP2003309493A - Method, device and program for reducing echo - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To propose an echo reducing method for suppressing only echo components from a voice acquisition signal superimposed with the voice of a speaker around. <P>SOLUTION: In a speech communication system with amplified voices, an M channel reproduced signal, and the power spectrum and cross spectrum of a voice acquisition signal are calculated, the coherence between the M channel reproduced signal and the voice acquisition signal of at least one channel or more is calculated, the rate of echo components occupying the voice acquisition signal is estimated from them for each frequency band, an echo suppression gain is calculated from the rate, and the short time spectrum of the voice acquisition signal is multiplied by the echo suppression gain to suppress the echo. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a reverberation reducing method, a reverberation reducing device, and a reverberation reducing program for suppressing acoustic echo that causes howling in a loud voice communication system.

[0002]

2. Description of the Related Art In a voice call system, a received voice is voiced from a speaker and picked up by a microphone to generate an acoustic echo. If the voice is transmitted as it is, problems such as a call disturbance and discomfort occur. When the loop gain of the closed loop formed including the voice communication system to the ground is larger than 1, the acoustic echo causes howling and makes the communication impossible. In order to overcome such a problem of the voice communication system and realize a natural communication environment, an echo canceller (echo canceling device) is used to cancel the echo caused by the acoustic sneak from the speaker to the microphone. Echo canceller
Echo cancellation is performed by estimating the echo path from the speaker reproduction signal and the sound collection signal, predicting the acoustic echo signal, and subtracting the predicted echo signal from the sound collection signal.

An acoustic echo canceller composed of an M-channel reproduction system and a 1-channel sound collection system eliminates acoustic echoes by the structure shown in FIG. Receiver terminal 1 _m (m =
1. ． ． The received signal from M) is the speaker 2 _m (m =
1. ． ． M) is reproduced as an acoustic signal, and circulates to the microphone 3 via the acoustic echo path. At the same time, the predicted echo signal is input to the predicted echo generation unit 41 to generate a predicted echo signal.
The subtractor 42 collects the sound pickup signal y (k) from the microphone 3.
And the predicted echo signal are obtained, and this residual signal is fed back to the echo path estimation unit 43. When there are N microphones, the configuration is such that N M-input 1-output type echo cancellers 4 as shown in FIG. 7 are arranged in parallel. Assuming that the impulse response of the acoustic echo path from the speaker 2 _m to the microphone is h _m (k) and the length thereof is L, when the number M of receiving channels is M = 1, between the input signal and the picked-up signal,

[Equation 3] And the impulse response and the input signal

[Equation 4] When the vectorization is performed as described above, the relationship between the input signal and the collected sound signal is simply described as follows.

[0004]

[Equation 5] Even when the number of receiving channels M ≧ 2, the impulse response and the input signal are

[Equation 6] By vectorizing as described above, the relationship between the input signal and the picked-up signal can be described in the same manner as in the case where the number of reception channels M = 1. Inside the echo canceller 4, the predicted echo generator 41
A predictive echo signal is generated by the adaptive filter for predictive echo generation so that the difference between the picked-up signal and the predicted echo signal becomes small based on the difference e (k) from the actual picked-up signal and the past received signal. Coefficients are updated sequentially. When the NLMS method is used to update the adaptive filter coefficient, the adaptive filter is

[Equation 7] Updated by However, μ is a step size set to a fixed value of 0 to 1 to stabilize the estimation, and e
(k) is a residual signal obtained by subtracting the predicted echo signal by the adaptive filter from the collected sound signal y (k). With the adaptive filter converged, an echo signal whose amplitude and phase match in each frequency band can be predicted, and the echo can be sufficiently canceled.

[0005]

However, in a situation where the echo canceller is actually used, it is not always possible to sufficiently cancel the echo. A large amount of information is required until a predicted echo signal whose amplitude and phase match in each frequency band is generated. When voice is input, it takes several seconds to converge even if a high-speed adaptive algorithm such as a projection algorithm is used. The echo path easily changes due to the movement of the human body, etc., and the residual echo increases for a few seconds immediately after the change.

The adaptive filter estimates the echo path on the assumption that only the acoustic echo signal generated by reproducing the far-end speaker's voice by the speaker is picked up and the near-end speaker's voice does not exist. Therefore, in the double-talk state in which the far-end speaker and the near-end speaker talk at the same time, the echo path estimation becomes unstable. To avoid this, usually by detecting the double talk state and setting the step size μ to 0,
Echo path estimation is stopped. However, when the estimation of the echo path is insufficient, the residual signal includes both the residual echo signal and the transmission signal, and thus the double-talk detection rate is reduced. If the double-talk determination is erroneous and the adaptive filter is updated in a state where the picked-up signal includes a signal other than the echo, the estimation error of the echo path increases and the residual echo signal increases.

Further, in the multi-channel echo canceller, the estimated echo path and the true echo path do not always match even when the echo is canceled because the received signal has a high inter-channel correlation. , Literature
MM Sondhi, DR Morgan, and JL Hall, "Stere
o-phonic Acoustic Echo Cancellation -An Overviewof
theFundamental Problem, ”IEEE Signal Processing L
It is analyzed in detail in etters, vol.2, no.8, pp.148-151 (1995). When the estimated echo path and the true echo path do not coincide with each other, the acoustic echo suddenly ceases to be canceled at the moment when the talkers take turns and the cross-correlation between channels of the received signal changes.

If there is a deviation between the estimated echo path and the estimated echo path in this way, residual echo occurs and the speech quality deteriorates. Therefore, it is assumed that there is a method for estimating the ratio of the echo component in the sound pickup signal in each frequency band as needed. Then, it becomes possible to attenuate the amplitude of the picked-up signal by an amount corresponding to the acoustic echo y _E (k) in each frequency band. Since the overlap between the voice spectrum of the near-end speaker and the echo spectrum is small in many cases, the spectrum of the acoustic echo can be suppressed while maintaining the near-end voice spectrum as much as possible, and it is expected that the speech quality will be improved.

In this process, the echo phase is not estimated in each frequency band. Therefore, compared to an adaptive filter that tries to match the amplitude / phase of the predicted echo with the amplitude / phase of the picked-up echo, less information is needed for estimation and the response to echo path changes etc. is improved. there is a possibility. If the ratio of the echo component in the collected signal can be obtained regardless of the presence or absence of the near-end speaker's voice,
Echo can be suppressed without being affected by the result of double-talk detection. However, it has been considered difficult to estimate the echo component ratio in the collected sound signal. This is because the near-end speaker's voice is superimposed on the echo signal to form a sound collection signal, and only the echo signal cannot be separated and extracted from the sound collection signal. If the voice signal power of the near-end speaker is constant and its level is known in advance, the power ratio of the echo component in the sound pickup signal can be calculated.
However, in a normal case, the voice level fluctuates from moment to moment and cannot be considered constant.

[0010]

According to the present invention, M speakers (M is an integer of 2 or more) and N microphones (N is 1) are used.
The above integers) are placed in a common sound field, and the
In a loudspeaker communication system that reproduces a channel reception signal and processes a sound pickup signal from a microphone into a transmission signal, obtains an M channel reproduction signal, a power spectrum and a cross spectrum of the sound collection signal, and obtains at least the M channel reproduction signal. Obtain the coherence of the picked-up signal of more than one channel, estimate the ratio of the echo component in the picked-up signal for each frequency band from these, calculate the echo suppression gain from this ratio, and echo in the short-time spectrum of the picked-up signal. We propose an echo reduction method that suppresses echoes by multiplying the suppression gain.

According to the present invention, further, in the echo reduction method, the first coherence γ is calculated from the short-time spectrum of the first channel reproduction signal and the short-time spectrum of the sound collection signal y (k).
² _1y (f) is obtained, and the short-time spectrum of the signal obtained by removing the correlation component between the m-th channel reproduced signal (m ≧ 2) and the 1st to (m-1) th channel reproduced signals and the picked-up signal y (k) From 1st to m-
The m-th coherence γ ² _{my (m-1)} (f) from the short-time spectrum of the signal from which the correlation component with the 1-channel playback signal has been removed
From the 1st to Mth coherences,

[Equation 8] We propose a method for reducing echoes required in.

Further, in the present invention, in any one of the above echo reduction methods, the power ratio of the received echo component in the sound pickup signal designated for each frequency band is γ ² (f),

[Equation 9] We propose a method for reducing echoes set to. In any one of the echo reduction methods according to the present invention, the signal output to the speaker is input to the pseudo echo path to generate the pseudo echo signal, and the pseudo echo signal is subtracted from the sound pickup signal from the microphone. We propose an echo reduction method that suppresses echoes for a picked-up signal that has undergone echo cancellation processing that updates the pseudo echo path based on the residual signal. Further, in the present invention, M speakers (M is an integer of 2 or more) and N microphones (N is an integer of 1 or more) arranged in a common sound field are connected, and the powers of the M channel reproduction signal and the sound collection signal are connected. The means for obtaining the spectrum and the means for obtaining the cross spectrum, and the coherence of the M-channel reproduction signal and the picked-up signal from the information of the power spectrum and the information of the crossed spectrum, determine the ratio of the echo component in the picked-up signal for each frequency band. An echo reduction apparatus is proposed which includes a means for estimating, a means for calculating an echo suppression gain from this ratio, and a means for multiplying a short-time spectrum of a sound pickup signal by the echo suppression gain.

According to the present invention, further, in the echo reduction device, means for obtaining a first coherence from the short-time spectrum of the first channel reproduction signal and the short-time spectrum of the sound pickup signal, and the m-th channel reproduction signal (m ≧ 2) ) From the first
From the short-time spectrum of the signal from which the correlation component with the m-1th channel reproduction signal has been removed and from the short-time spectrum of the signal from which the correlation component with the 1st to m-1th channel reproduction signal has been removed from the collected signal And a means for determining the coherence of the echo reduction device. According to the present invention, further, in any of the echo reduction devices, means for generating a pseudo echo signal by inputting a reproduction signal output to a speaker into a pseudo echo path, and the pseudo echo signal from a sound pickup signal from a microphone. We propose an echo reduction device that targets a sound pickup signal that has passed through an echo cancellation processing unit that includes a subtraction unit and a unit that updates a pseudo echo path based on the residual signal.
The present invention further proposes an echo reduction program that executes any of the echo reduction methods by a computer.

[0014] the purpose of estimating the proportion of echo component occupying the action collected sound signal,
It is possible to use the magnitude squared value of the complex function obtained by normalizing the coherence, that is, the cross spectrum with the power spectrum. If the acoustic echo signal is y _E (k) and the non-echo signal such as the voice of the near-end speaker is y _I (k), the picked-up signal is

[Equation 10] become. In the case of the number of reproduction channels M = 1, the coherence between the reproduction signal and the picked-up signal is

[Equation 11] Is defined by However, S _xx (f): power spectrum of reproduced signal S _yy (f): power spectrum of collected signal S _xy (f): cross spectrum of reproduced signal and collected signal. Normally, the playback signal x (k) and the signal y other than the echo are
_I (k), and echo signal y _E (k) and non-echo signal y _I (k)
Can be considered uncorrelated, so

[Equation 12] Has been established. Also, from the reproduced signal x (k) to the echo signal y _E
When the transfer characteristic to (k) can be regarded as almost constant, the cross spectrum of two signals

[Equation 13] Has been established. From this, the playback signal x (k) and the picked-up signal y (k)
The coherence of

[Equation 14] Meets

According to this equation, this coherence is
It is the ratio of the component having a correlation with the reproduction signal to the power spectrum of the collected sound signal. That is, the coherence between the reproduced signal and the collected sound signal represents the power ratio of the echo component in the collected sound signal. Details of coherence are described in, for example, "Spectrum Analysis" published by Hino and Asakura Shoten, and details of analysis using coherence are described in, for example, "Signal Processing" by Morishita and Obata. Therefore, in terms of signal power, the ratio of signals other than echoes in the collected sound signal is

[Equation 15] Required by. In each frequency band, Y (f), which is the sound pickup signal y (k) framed and transformed into the frequency domain by Fourier transform,

[Equation 16] The echo component can be suppressed by processing as described above. This processing result Z (f) is transformed into the time domain to obtain the signal z (k) in which the echo component is suppressed. Estimated by soft judgment
(Soft-decision estimation), estimation by least square error
(Minimum Mean Square Error estimation), maximum likelihood estimation method
When (Maximum Likelihood estimation) is used, the ratio of the signal other than the echo to the picked-up signal and the echo suppression gain can be obtained from γ ² (f) by a plurality of methods. For each estimation method, see P. Scalart and JV.
Filho, "Speech Enhancement based on a priori sign
al tonoise estimation, "Proc. ICASSP96, pp. 629-63
Details in 2 (1996).

Similarly, for any number of reproduction channels, coherence can be used as a ratio of echo components in a sound pickup signal. For example, when the number of reproduction channels M = 2, the ratio of the echo component in the sound pickup signal is

[Equation 17] X ₂ (k) where the correlation component between γ ² _1y (f): x ₁ (k) and the coherence γ ² _{2y (1)} (f): x ₁ (k) of y (k) is removed. And the coherence obtained from y (k). Regarding the coherence of the above-mentioned multi-input 1-output system, see
JS Bendat and AG Piersol, "Engineering Applica
tions of Correlation and Spectral Analysis, "John
Familiar with Wiley & Sons (1980).

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION The echo reduction method according to the present invention is
This is realized by the signal flow of the acoustic echo suppression unit 6 in FIG. The acoustic echo suppressor 6 shown in FIG. 1 can be used in place of the echo canceller 4 in the signal flow diagram (FIG. 7) of the voice communication system. Hereinafter, a case will be described in which each signal is divided into blocks for each L / D sample with 1 frame = 2L samples. Step 1 Time domain-frequency domain transforming unit (hereinafter referred to as TF transforming unit) TF transforming units 61 _{I to} 61 _M play signals x _I (k) to x _M (k), and TF transforming unit 62 collects sound signals y. (k) respectively

[Equation 18] To the frequency domain coefficient.

Step 2 The echo component ratio estimation unit 63 uses the M at time k = j L / D.
The short-time spectrums of the channel reproduction signal and the sound collection signal are input, and the ratio of the echo component in the sound collection signal is estimated for each frequency band according to the signal flow of FIG. Correlation remover
631 _m (m = 2 to M) is the first to the first m-th channel reproduction signal x _m (k)
A short-time spectrum X _{m (m-1)} (f, j) of the signal from which the correlation component with the m-1th channel reproduction signal is removed is obtained. However, the input is the reproduction signal of the first channel and the reproduction signals of the 2nd to m-1th channels that have already undergone the correlation component removal processing, and the actual processing is described by the following equation.

[Formula 19] Here, ε [] means to take a time average. The time averaging process is, for example,

[Equation 20] As described above, there is a method of using the processing result of one frame before and the smoothing constant β having a value of 0 to 1. Also, the correlation removing unit 63
For 2 _m (m = 2 to M), a short-time spectrum of the signal obtained by removing the correlation component with the reproduced signals of the 1st to m-1th channels from the collected signal y (k) is obtained. The process is

[Equation 21] And is almost the same as the above correlation removing unit 631 _m . Then, the coherence calculation unit 633 ₁ obtains the coherence of the first channel reproduced signal and the collected sound signal, the coherence calculator
At 633 _m (m = 2 to M), coherence is obtained from the short-time spectrum of the reproduced signal and the picked-up signal that have undergone the correlation removal processing.

[Equation 22] Using these coherences, the echo component ratio calculation unit 63
4 calculates the ratio of the echo component in the picked-up signal.

[Equation 23]

Step 3 For each frequency band, the attenuation ratio calculation unit 64 obtains the amplitude attenuation rate from the echo component ratio in the sound pickup signal, and the multiplier 65 attenuates the amplitude of the sound pickup signal. As an example of the amplitude attenuation rate obtained from the echo component ratio, the attenuation rate as shown in the following equation can be considered.

[Equation 24] This suppresses the echo.

[Equation 25]

Step 4 The processing result in the frequency domain is obtained by the FT converter 66.

[Equation 26] Inverse FFT transformation is used to transform into a block signal in the time domain. From this block signal, for example,

[Equation 27] A signal after echo suppression processing may be obtained by cutting out a part of the frame as described above, or the signal after echo suppression processing may be obtained by performing window processing on a plurality of frames and combining overlapping sections. .

[Second Embodiment] FIG. 3 shows another acoustic echo suppressing method of the present invention. In the configuration shown in FIG. 3, the acoustic echo suppression method of the present invention is combined with the acoustic echo cancellation method using an adaptive filter. Here, the acoustic echo suppression processing performed by the acoustic echo suppression unit 6 is applied to the signal that has undergone the echo cancellation processing by the echo cancellation unit 4 instead of the collected signal. It is known that when noise is picked up by a microphone at the same time as acoustic echo, the accuracy of echo path estimation reaches a ceiling due to the effect of noise, and acoustic echo continues to remain audibly. Even in such an acoustically severe case, it is possible to maintain a high communication quality by combining the acoustic echo suppression method with the acoustic echo cancellation method using the adaptive filter.

[Embodiment 3] In the third embodiment, a method of performing echo suppression based on coherence is described in the literature: Emura, Haneda, "Frequency-domain stereo adaptive algorithm with additional signal enhancement", Acoustical Society of Japan 2001 Autumn Research. Recital, p
The case of combining with the multi-channel adaptive algorithm proposed in p. 537-538 (2001) is explained. This adaptive algorithm obtains the correction vector of the adaptive filter from the correction signal instead of the input signal. Therefore, in addition to the M channel received signal (M is an integer of 2 or more), the M channel echo canceling unit 7 of FIG.
The M-channel additional signals generated by 8 ₁ to 8 _M are also input. The correlation variation process is a means generally used for improving the echo path estimation performance of the multi-channel echo canceller. The acoustic echo suppression processing executed by the acoustic echo suppression unit 6 is performed on the signal subjected to the echo cancellation processing by the M channel echo cancellation unit 7.

In the M channel echo canceling section 7 of FIG. 4, the coefficient of the adaptive filter is updated according to steps 1 to 6 below. Then, the acoustic echo suppression unit 6 performs echo suppression processing according to steps 7 to 9 below. Step 1 From the received signal u _m (k) of each channel and the additional signal g _m (u _m (k)) for correlation fluctuation processing, the reproduced signal x _m (k) and the correction signal v
_m (k)

[Equation 28] Generated by. However, a is a value greater than 0 and less than or equal to 1. Then, for each L / D sample, block into a signal vector of length 2L, and use FFT

[Equation 29] To the frequency domain. The function diag () transforms a vector into a matrix whose elements are diagonal elements.

[Equation 30] Step two

[Equation 31]

Step 3

[Equation 32] Step 4

[Expression 33] Step 5

[Equation 34] Step 6

[Equation 35] Then, the adaptive filter of each channel is updated by the following equation.

[Equation 36] However, μ is a step size that takes a value of 0 to 1. Step 7 Use FFT to find the vector consisting of the residual signal at time jL / D

[Equation 37] To the frequency domain.

Step 8 In the correlation removing unit of the echo component ratio estimating unit 63, time k = j L /
With the short-time spectrum of the M channel reproduction signal and the residual signal in D as input, the correlation component between the mth channel reproduction signal x _m (k) and the 1st to m-1th channel reproduction signals is removed according to the flow of FIG. The short-term spectrum of the signal

[Equation 38] Ask by. In addition, the short-term spectrum of the signal obtained by removing the correlation component with the signal 1 to m-1th channel reproduction signal from the residual signal

[Formula 39] Ask by. However, ε [] means to take the time average. For example, the time averaging process is

[Formula 40] As described above, there is a method of using the smoothing constant β (0 to 1) and the processing result of one frame before. And from the short-term spectra that have been de-correlated

[Formula 41] Seek coherence by. Further, the coherence between the reproduction signal of the first channel and the picked-up signal is also obtained. Then, using these coherences, the ratio of the echo component in the residual signal is obtained.

[Equation 42]

Step 9 From the ratio γ ² (f) of the echo components in the residual signal obtained in step 8, the amplitude attenuation rate is obtained for each frequency band,
It is applied to the residual signal in the frequency domain to suppress the residual echo. As an example of the method of obtaining the amplitude attenuation rate from the echo component ratio, the attenuation rate as shown in the following equation can be considered.

[Equation 43] Then, Es (j) is returned to the time domain by the inverse FFT transform,
A signal with suppressed residual echo is obtained.

[Equation 44]

"Demonstration Example of Effect" The result of actual numerical simulation of the method of Example 1 is shown in FIG.
As shown in FIG. In this numerical simulation, the number of input channels was M = 2 and the sampling frequency was set to 8 kHz. As the acoustic echo path, the room transfer function measured in a room with a reverberation time of 200 ms was truncated to 700 taps to generate an acoustic echo. The input signals of two channels were generated by simulating the voice stereo sound pickup of two far-end speakers using the measured indoor transfer function. The speakers are taking turns at t = 5.3s. This two-channel signal, as the inter-channel correlation fluctuation processing,
P. Eneroth, T. Gaensler, S. Gay and J. Benesty, "Stu
dies of a Wideband Stereophonic Acoustic Echo Canc
The half-wave rectification method proposed by eller, "Proc. IWAENC, pp. 207-210 (1999). was applied with an additional gain of 0.25.

Using this signal, the conventional method of canceling the echo by the adaptive filter and the echo suppressing method of the first embodiment are compared. As an adaptive algorithm, reference D. Mansour a
nd AH Gray, "Unconstrained Frequency-Domain Ada
ptive Filter, "IEEE Trans.onAcoust., Speech, Signal
Processing, vol. ASSP-30, No. 5, pp. 726-734 (198
An algorithm that extends the proposed algorithm in 2) to multiple channels is used. The number of adaptive filter taps per channel is L = 512, and the adaptive filter has 256 samples,
D = 2 is set so that it is updated every 32 ms. The step size was set to μ = 0.3.

Sound collection signal = acoustic echo signal + speech signal of near-end speaker, signal after echo cancellation processing by adaptive filter,
And the voice signal of the near-end speaker is shown in FIG. Interval t = 0.3
In ~ 1s, the residual echo is conspicuous because the estimation by the adaptive filter is insufficient. Immediately after the speaker change (t = 5.3
In s), the residual echo suddenly increases. Similarly,
FIG. 6 shows the sound collection signal = acoustic echo signal + speech signal of the near-end speaker, the signal after the echo suppression processing of the present invention, and the speech signal of the near-end speaker. Β = 0.25 was used as the smoothing constant for coherence estimation. According to this graph, t = 0.3 ~ 1
The echo is well suppressed even in the section of s. Immediately after the change of the speaker (t = 5.3s), there is no sudden increase in echo like the adaptive filter. Also, in the interval from t = 7.5 to 9 s,
Although the amplitude is attenuated, the general shape of the near-end speaker's voice is maintained. Thus, it was found that the proposed echo suppression method has good responsiveness to changes in acoustic conditions.

The echo reduction method according to the present invention described above is realized by executing a program described by a computer-readable code on a computer. The program is downloaded to a computer via a recording medium such as a CD-ROM or a communication line, installed, and executed by an arithmetic means such as a CPU.

[0031]

According to the present invention, the ratio of the echo component occupied in the collected sound signal is estimated for each frequency band, and the amplitude of the collected sound signal is attenuated by an amount corresponding to the echo. This makes it possible to suppress the acoustic echo even if the acoustic conditions suddenly change due to a change in speakers.

[Brief description of drawings]

FIG. 1 is a block diagram for explaining an example of an acoustic echo suppression unit that executes an echo reduction method of the present invention.

FIG. 2 is a block diagram for explaining the inside of an echo component ratio estimation unit used in the acoustic echo suppression unit shown in FIG.

FIG. 3 is a block diagram for explaining an embodiment in which the echo reduction method according to the present invention and a conventional echo cancellation method are combined.

FIG. 4 is a block diagram for explaining an embodiment in which the echo reduction method according to the present invention and a conventional M channel echo cancellation method are combined.

FIG. 5 is a graph for explaining an echo canceling effect according to a conventional technique.

FIG. 6 is a graph for explaining an echo suppressing effect by the echo reducing method according to the present invention.

FIG. 7 is a block diagram for explaining a conventional technique.

[Explanation of symbols]

6 Acoustic Echo Suppression Unit 64 Attenuation Ratio Calculation Unit 61 _{I to} 61 _M , 62 TF Conversion Unit 65
Multiplier 63 Echo component ratio estimator 66 FT converter

   ─────────────────────────────────────────────────── ─── Continued front page    F term (reference) 5K027 BB03 DD07 DD10 HH03                 5K046 HH01 HH24 HH25 HH31 HH78                       HH79

Claims (8)

[Claims] 1. M speakers (M is an integer of 2 or more) and N microphones (N is an integer of 1 or more) are arranged in a common sound field to reproduce an M channel reception signal from the speakers, In a voice communication system that processes a picked-up signal into a transmission signal, the power spectrum and cross spectrum of the M-channel reproduced signal, the picked-up signal are obtained, and the coherence between the M-channel reproduced signal and the picked-up signal of at least one channel is determined. Seeking,
Estimate the ratio of the echo component occupying the collected signal for each frequency band, calculate the echo suppression gain from this ratio, and multiply the short-time spectrum of the collected signal by the echo suppression gain to suppress the echo. An echo reduction method characterized by the above. 2. The echo reduction method according to claim 1, wherein
Short-term spectrum of the 1st channel playback signal and the collected sound signal y
From the short-time spectrum of (k), the first coherence γ
² _1y (f) is _calculated, and the short-time spectrum of the signal obtained by removing the correlation component from the m-th channel reproduced signal (m ≧ 2) with the 1st to m-1th channel reproduced signals and the picked-up signal y (k) From the short-time spectrum of the signal from which the correlation component with the _{1st to (m-1) th} channel reproduced signals has been removed, the mth coherence γ ² _{my (m-1)} (f) is obtained, and the 1st to Mth coherences are obtained. The ratio of the echo component in the picked up signal is Reverberation reduction method characterized by obtaining in. 3. The echo reduction method according to claim 1, wherein the power ratio of the received echo component in the sound pickup signal estimated for each frequency band is γ ² (f), and Reverberation reduction method characterized by setting to. 4. The echo reduction method according to claim 1, 2, or 3, wherein a signal output to a speaker is input to a pseudo echo path to generate a pseudo echo signal, and a sound pickup signal from a microphone is collected. The echo reduction method is characterized in that the echo is suppressed for a sound pickup signal subjected to the echo cancellation processing of updating the pseudo echo path based on the residual signal of the pseudo echo signal. 5. The power of an M channel reproduction signal and a picked-up signal connected to M speakers (M is an integer of 2 or more) and N microphones (N is an integer of 1 or more) arranged in a common sound field. A means for obtaining a spectrum, a means for obtaining a cross spectrum, a coherence of an M channel reproduction signal and at least one channel of a sound pickup signal from power spectrum information and cross spectrum information, and from these, a sound pickup signal for each frequency band is obtained. An echo reduction apparatus comprising: a means for estimating a ratio of occupied echo components; a means for calculating an echo suppression gain from this ratio; and a means for multiplying a short time spectrum of a sound collection signal by the echo suppression gain. 6. The echo reduction device according to claim 5, wherein the first coherence is obtained from the short-time spectrum of the first channel reproduction signal and the short-time spectrum of the sound pickup signal, and the m-th channel reproduction signal (m ≧). 2) from the short-time spectrum of the signal from which the correlation component with the 1st to m-1th channel reproduction signals has been removed, and the signal from which the correlation component with the 1st to m-1th channel reproduction signals has been removed from the collected signal A means for obtaining an m-th coherence from a short-time spectrum, and an echo reduction device comprising: 7. The echo reduction device according to claim 5, wherein the reproduction signal output to the speaker is input to the pseudo echo path to generate a pseudo echo signal, and sound is collected from the microphone. An echo reduction device, wherein the picked-up sound signal passed through the echo cancellation processing means by means for subtracting the pseudo echo signal from the signal and means for updating the pseudo echo path based on the residual signal is targeted. . 8. An echo reduction program for executing the echo reduction method according to claim 1 by a computer.