JP2003250193A - Echo elimination method, device for executing the method, program and recording medium therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an echo elimination method for strongly estimating the echo route of an adaptive filter even under the conditions that interference signals other than echoes are present, and to provide a device and a program. <P>SOLUTION: In the echo elimination method for reproducing M channel reception signals (M: an integer of ≥2) and subtracting pseudo echo signals from sound pick-up signals, the pseudo echo signals are generated by adding convolution signals generated by the convolution operation of adaptive filter coefficients corresponding to respective channels to the M channel reception signals over the respective channels and the ratio of echo components occupying residual signals obtained by subtracting the pseudo echo signals from the sound pick-up signals is obtained. A correction vector composed of the product of the reception signals of the respective channels corresponding to the residual signals is corrected by the ratio of the echo components occupying the residual signals and updated vectors are added. Thus, the frequency domain coefficient of the adaptive filter coefficient is updated. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to an echo canceling method,
The present invention relates to an apparatus, a program, and a recording medium thereof, and more particularly, to an echo canceling method, an apparatus, a program, and a recording medium for canceling an echo that sometimes causes howling in an acoustic communication apparatus such as a voice communication apparatus.

[0002]

2. Description of the Related Art In a loudspeaker communication device, there is a problem of a reverberation that occurs when a received voice is amplified by a speaker and sneaks into a microphone. If the loop gain of the closed loop of the loudspeaker communication apparatus interconnected via the communication line is larger than 1, howling is caused to make the communication impossible. Even when the loop gain is smaller than 1, the echo causes an obstacle to the call and gives an unpleasant feeling. In order to realize a more natural call environment, it is necessary to eliminate the echo caused by the acoustic sneak from the speaker to the microphone.

Referring to FIG. 1, the echo canceller is connected to an M-channel reproduction system and a 1-channel sound pickup system to cancel echoes. Here, the reception signal input from the reception terminal 1 _m (m = 1 to M) is reproduced as an acoustic signal in the speaker 2 _m (m = 1 to M), and the echo path h _m
Go around the microphone 3 via (m = 1 to M).
The echo canceller 4 connected between the reception terminal 1 _m and the transmission terminal 5 cancels the echo. This echo canceller 4 receives M input 1
It consists of an output adaptive filter. When there are N microphones 3, the M input 1 output adaptive filter shown in FIG.
It is arranged in parallel.

The structure of the echo canceller 4 will be described with reference to FIG. Each received signal is input to the predictive echo generator 41 to generate a predictive echo signal, and the difference between the predicted echo signal and the sound pickup signal input from the microphone 3 is taken in the subtractor 42, and the residual signal is obtained. e (k) is the echo path estimation unit 43
Be fed back to. The input signal to the predicted echo generation unit 41 is x _m (k), the sound pickup signal picked up by the microphone 3 is y (k), the impulse response of the echo path from the speaker 2 _m to the microphone 3 is h _m , If the length is L, and the number of receiving channels M = 1, the input signal x _m (k)
Between the sound pickup signal y (k) and

[0005]

[Equation 6]

By vectorizing as described above, the input signal x
The relationship between (k) and the picked-up signal y (k) can be described in the same manner as in the case where the number of receiving channels M = 1. In the echo canceller 4, a predicted echo signal is generated by the predicted echo generator 41, and a difference e between the predicted echo signal y (k) and the actual sound pickup signal y (k) is generated.
(k) and the input signal x _m (k) in the past so that the residual signal e (k), which is the difference between the picked-up signal y (k) and the predicted echo signal, becomes small so that the predicted echo signal is generated. The filter coefficients are updated sequentially. Here, a case will be described where the adaptive filter coefficient updating method is the NLMS method. The residual signal e (k) obtained by subtracting the predicted echo signal predicted by the adaptive filter from the actual sound pickup signal y (k) is

[0007]

[Equation 7]

Update by However, μ is a step size set to a fixed value of 0 to 1 in order to stabilize the estimation. In this adaptive filter updating method, it is premised that the picked-up signal y (k) is the picked-up echo only. However, when the voice communication device is actually used, the sound pickup signal y (k) naturally includes signals other than reverberation such as speech and noise. Here, a reverberation signal is y _E (k), a signal other than reverberation such as speech and noise is a disturbance signal y _I (k), and the sound pickup signal y (k) is y (k) = y _E (k). Let + y _I (k). At this time, the adaptive filter update formula of the NLMS method is

[0009]

[Equation 8]

It is corrected in the direction of. However, ε [·] means to take an average. This second term represents the deviation from the ideal correction direction, and it can be seen that speech and noise act as interference signals. Interfering signal y in the picked up signal y (k)
_{When I} (k) is included, the adaptive filter coefficient is erroneously updated by this amount, so the step size μ
Noise is generated according to the value of, and sometimes the adaptive filter is diverged. In order to avoid divergence, it is necessary to make the step size μ sufficiently small, but in reality, select an unnecessarily small μ, or use a μ that is small enough not to diverge to avoid interference due to acoustic interference other than reverberation. Accurate correction is allowed with a certain probability, which leads to a decrease in convergence speed.

References A. Mader, H. Puder, GU Schmidt, “Ste
p-size control for acoustic echocancellation filte
rs-an overview, ”Signal Processing, 80, pp.1697-1719
(2000) shows a method for deriving an optimum step size μ in such a situation. According to this, the residual echo signal, which is the difference between the echo and the predicted echo,

[0012]

[Equation 9]

Is calculated by According to this formula, the step size μ increases as the interference signal power ε [y _I ² (k)] increases.
Is set small, the influence of the interference signal y _I (k) on the adaptive filter estimation is reduced.

[0014]

However, it was not possible to update the adaptive filter by directly obtaining this optimum step size μ in an actual environment. That is, from the residual signal in which the disturbing signal is superimposed on the residual echo signal e _E (k),
This is because it is not possible to extract only the residual echo signal e _E (k). In addition, the echo canceller is originally a speaker 2 _m.
Is used to eliminate the echo while estimating the unknown echo path h from the microphone to the microphone 3, so the relational expression of e _E (k) = (h (k) −h ^{^} (k)) ^T x (k) This is because the residual echo signal cannot be obtained from

Let us assume that the power ε [y of the interference signal y _I (k) is
_{If I} ² (k)] is constant and its level is known in advance, the optimum step size μ can be calculated.
However, normally, the level of the noise signal is not always constant, and the level of the transmission signal varies from moment to moment. In the above situation, in order to update the adaptive filter using the optimum step size μ, it is necessary to estimate the ratio of the reverberation component in the residual signal. An object of the present invention is to solve the above-mentioned problems in multi-channel acoustic communication by obtaining the ratio of the reverberation component occupying in the residual signal from the residual signal or the picked-up signal and updating the adaptive filter coefficient using this information. To provide an echo canceling method, device, program, and recording medium therefor.

[0016]

According to the present invention, M speakers (M is an integer of 2 or more) and N microphones (N) are used.
Are integers greater than or equal to 1) are arranged in a common sound field, reproduce the M channel signal from the speaker, and input the M channel reproduction signal to each M input 1 output adaptive filter corresponding to each microphone to predict the echo signal. , In a multi-channel acoustic communication system in which the adaptive filter coefficient is updated so as to reduce the residual signal obtained by subtracting the adaptive filter output signal from the sound pickup signal from the microphone, the ratio of the reverberation component to the residual signal is used. And an echo canceling method for updating the adaptive filter coefficient. It is also possible to configure an echo canceling method for updating the adaptive filter coefficient by using the ratio of the echo component in the picked-up signal instead of the residual signal. As a result, echo cancellation and echo path estimation by the adaptive filter are stable even in a situation where the collected signal includes signals other than echo.

Further, the M-channel reproduction signal is converted into the frequency domain for each short time interval, multiplied by the adaptive filter coefficient in the frequency domain, converted into the time domain to predict the echo signal, and the echo predicted from the picked-up signal. The residual signal obtained by subtracting the signal is converted into the frequency domain for each short time interval, and the ratio of the reverberation component to the target signal for each frequency band is obtained from the short time spectrum of the reproduced signal and the target signal. The correction vector obtained by multiplying the residual signal and the reproduction signal for each frequency component in the frequency domain is the ratio of the reverberation component to the target signal,
And, an echo canceling method for updating the adaptive filter coefficient by correcting each frequency band based on the information of the input signal and the correction signal is constructed. By handling the adaptive filter coefficient in the frequency domain, it is possible to significantly reduce the total calculation amount while stabilizing the echo cancellation and the echo path estimation in the situation where the collected signal includes signals other than the echo.

Further, the M-channel reception signal is processed to generate an M-channel additional signal whose inter-channel correlation can be regarded as almost uncorrelated, and added to the reception signal to be a reproduction signal, which is converted into a frequency domain for each short time section. After that, the adaptive filter coefficient in the frequency domain is multiplied and then converted into the time domain to predict the echo signal, and the residual signal between the picked-up signal and the predicted echo signal is converted into the frequency domain for each short-term interval, and the reproduced signal And the short-term spectrum of the target signal, obtain the ratio of the reverberation component in the target signal for each frequency band, and add the M channel additional signal to the M channel additional signal a times (a is a value between 0 and 1) for correction. A signal is generated, the correction signal is converted into the frequency domain for each short time interval, and the correction vector obtained by multiplying the residual signal and the correction signal for each frequency component in the frequency domain An echo canceling method is constructed in which correction is performed for each frequency band based on the ratio and information of the input signal and the correction signal, and the adaptive filter coefficient is updated with the corrected correction vector. As a result, echo cancellation and echo path estimation can be stabilized in a situation where the collected signal includes a signal other than the echo, and the echo path estimation can be speeded up while significantly reducing the total calculation amount.

Further, from the m-th channel reproduction signal,
The short-term spectrum of the signal from which the correlation component with the m-1 channel reproduction signal has been removed is obtained, and from the target signal,
An echo canceling method is obtained in which a short-time spectrum of a signal from which a correlation component with a channel reproduction signal is removed is obtained, and the coherence obtained from these short-time spectra is used to obtain a ratio of an echo component in a target signal. With such an estimation method, it is possible to reliably estimate the ratio of the reverberation component to the residual signal or the picked-up signal even in a situation where the reverberation other than the reverberation contained in the reproduced signal and the picked-up signal fluctuates every moment.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION A coherence, that is, a complex spectrum obtained by normalizing a cross spectrum with a power spectrum for the purpose of estimating a ratio of an echo component occupying the target signal when a residual signal or a sound pickup signal is set as the target signal. The magnitude squared value of the function can be used. Hereinafter, the case where the residual signal is the target signal will be described. Number of input channels is M
For the monaural echo canceller with = 1, the power spectra of the input signal x (k) to the adaptive filter and the residual signal e (k) are S _xx (f) and S _ee (f), and the cross spectrum is S _xe ( f)
And the coherence is

[0021]

[Equation 10]

Is calculated by Usually, the input signal x (k) and the disturbing signal y _I (k), and the residual reverberation signal e _E (k) and the disturbing signal y _I (k) can be regarded as uncorrelated.

[0023]

[Equation 11]

[0024] is satisfied. According to this equation, coherence γ ² (f) is the ratio of the component correlated with the input signal spectrum to the power spectrum of the residual signal e (k). That is, the coherence between the input signal x (k) and the residual signal e (k) represents the power ratio of the reverberant component in the residual signal e (k), that is, the residual echo signal e _E (k). Coherence is described in detail in, for example, Hino, published by Asakura Shoten, "Spectral Analysis," and analysis using coherence is described in detail in, for example, Morishita, Obata, and the Institute of Instrument and Control Engineers, "Signal Processing." There is.

Each power spectrum and cross spectrum
Is the input signal x (k) and the residual echo signal e _E(k) is separated by 2L
Short-time spectrum X (f), E obtained by Fourier transform
(f) (f = 1, ..., 2L) and time average ε [•]
From

[0026]

[Equation 12]

It is calculated as follows. Although the residual echo signal e _E (k) and the disturbing signal y _I (k) cannot be separated from the residual signal e (k), it is possible to obtain the optimum step size μ by performing this coherence analysis. become.

[0028]

[Equation 13]

[0029]

[Equation 14]

X _{m · (m−1)!} (F): signal x _m (k) to signal x ₁
(k), ..., Short-time spectrum of the signal from which the correlation component with x _(m-1) (k) has been removed, and E. _(m-1)! (f): signal e (k ) From the signal x ₁ (k), ...
It is the coherence of the short-time spectrum of the signal from which the correlation component with x _(m-1) (k) has been removed. Number of channels M = 2
Similarly to the case of, the short-time spectrum X _{m · (m-1)!} (F) after removing the correlation component is

[0031]

[Equation 15]

The above-described correlation component removal calculation is executed by the first correlation removal section 4321 _m and the second correlation removal section 4322 _m shown in FIG. The short time spectrum X _m (j, f) of the input signal and the spectrum of the signal whose correlation is removed are input to the first correlation removing unit 4321 _m to remove the correlation component, and the short time spectrum X _{m · (m -1)!} (J, f) is obtained. After removing the correlation component by inputting the spectrum X _{m · (m− 1)!} (J, f) of the echo-removed signal E (j, f) and the correlation removed to the second correlation removing unit 4322 _m To obtain the short-time spectrum E _{· (m−1)!}

Residual echo signal e_EPredicted value of (k) and input signal
coherence γ of x (k)^{^ 2}(f) for step size control
It is also possible to use it. As a method of predicting residual echo signals
For example, the echo signal y_EEach frequency component of (k) is multiplied by t (f)
There are possible ways to do this. As an example, t (f) = 0.1
If set, the residual echo signal power is set to the echo signal power.
-20 dB of the residual signal e (k)
Occupy residual echo signal e _ECorresponding to obtaining the ratio of (k)
To do. Of the M channel input signal and the residual signal e (k) described above.
Similar to the coherence calculation, M channel input signal x
₁(k) ・ ・ ・ ・ ・ x_M(k) and the coherence γ'of the picked-up signal y (k)
Echo signal in the residual signal when (f) is required
Ratio of ingredients γ^{^ 2}(f) is

[0034]

[Equation 16]

As in the above, it can be calculated from γ '(f). As a method of updating the adaptive filter, the above-mentioned NLM is used.
In addition to the method of performing processing of each sample in the time domain as in the S method, there is a block processing method of performing processing in each fixed section.
This is the document ERFerrara, “Fast Implementation of
LMS adaptive filters, ”IEEE Trans.Acoust., Speech
As already proposed in Signal Processing, vol.ASSP-28, pp.474-475 (1980), using FFT to handle adaptive filter coefficients in the frequency domain significantly reduces the total amount of calculation. can do. In this adaptive algorithm, the frequency domain adaptive filter coefficient vector H ^{^}
(J) is

[0036]

[Equation 17]

Embodiments of the present invention will be described below with reference to examples. Example 1 In Example 1, reference D. Mansour and AHGray, “U
nconstrained Frequency-Domain Adaptive Filter, ”IE
EE Trans.Acoust., Speech, Signal Processing, vol.ASSP
・ The case where the algorithm proposed in 30, No. 5, pp. 726-734 (1982) is extended to multi-channel and the step size control method based on coherence is applied will be described. This frequency domain adaptive algorithm prevents deterioration of the convergence characteristic of the adaptive filter even when a signal having a spectrum bias such as a received signal is input by the whitening process.

In the following description, the residual signal is the target signal,
Let L be the adaptive filter length and use the overlap-save method.
Process a 2L length signal vector for each L / D sample
You are dealing with the case. (Step 1) Input signal x_m(k) (m = 1, ..., M)
To a signal vector of length 2L for each L / D sample.
And converted into the frequency domain by FFT. X_m (J) = diag (FFT ([x_m(JL / D-2
L + 1), ..., x_m(JL / D)]^T),here,
(M = 1, ..., M) However, the function FFT (x) transforms the vector x by FFT.
And the vector x is the function diag (x)
It is converted into a matrix whose elements are diagonal elements. That is,
x = [x (1) ... x (2L)]^TWhen

[0039]

[Equation 18]

(Step 2) X _m (j) in the frequency domain
And the input signal vector is filtered for each channel by multiplying by the adaptive filter coefficient vector H ^{^} _m (j) in the frequency domain of the m-th channel. Inverse FFT of calculation result
The signal vector y ^{^} _m (j) in the time domain is obtained by processing. y ^{^} _m (j) = [0 _L I _L ] IFFT (X
_m (j) H ^{^} _m (j)) However, H ^{^} _m (j) is a complex vector having 2L elements, and the inverse FFT transform to extract the first L results in the impulse response of the adaptive filter. 0 _L is an L × L zero matrix, and I _L is an L × L identity matrix.

(Step 3) Signal vector y ^{^} _m (j)
Are added to obtain the vector y ^{^} (j) of the predicted echo signal. y ^{^} (j) = Σ ^M _{m = 1} y ^{^} _m (j) (Step 4) Sound signal vector y in the time domain
The residual signal vector E (j) is obtained from (j) and the predicted echo vector y ^{^} (j), and is transformed into the frequency domain by FFT.

[0042]

[Formula 19] (Step 5)

[0043]

[Equation 20]

[0044]

[Equation 21]

(Step 6) The residual signal and the input signal are processed in the frequency domain to obtain a correction vector dH ^{^} _m (j).

[0046]

[Equation 22]

However, each element of the matrix X ^* _m (k) is a complex conjugate of each element of the matrix _Xm (k). (Step 7) The matrix P (k) is

[0048]

[Equation 23]

It is the sum of power spectra of the input signal obtained by Where X ^* is the complex conjugate of the complex number X, and β is a smoothing constant for averaging in a short time, 0 <β <
Takes a value of 1. (Step 8) From the ratio γ ² (f) of the reverberation component in the residual signal obtained in Step 5,

[0050]

[Equation 24]

A matrix M (j) having diagonal elements of coherence γ ² (f) is obtained by However, μ ₀ is set to a fixed value between 0 and 1. Update the adaptive filter with the following formula. H ^{^} _m (j + 1) = H ^{^} _m (j) + M (j) P
(J) dH ^{^} _m (j) The step size is optimally controlled based on the ratio γ ² (f) of the reverberation component in the residual signal for each frequency band by multiplying by the matrix M (j). Multiplying the correction vector dH ^{^} _m (j) by the matrix P (j) corresponds to whitening processing of the input signal, and can improve the convergence characteristic of the adaptive filter when the input signal is a chromatic signal like speech. Are known.

The method of the first embodiment is carried out by the echo canceller 4 having the configuration shown in FIG. Input signal x ₁ (k) _...... x _M (k)
Is converted into blocks by the TF converters 411 _{1 to} 411 _{M as in} step 1 and converted into the frequency domain. The filter processing unit 412 ₁ ~412 _M and FT conversion unit 413 ₁
413 _M , the vector adder 414 calculates the vector y ^{^} (j) of the predicted echo signal in the time domain as in steps 2 and 3. The picked-up sound signal y (k) is divided into blocks by the blocking unit 45 so as not to be time-shifted from the input signal x (k), and the signal vector subtraction unit 42 calculates the predicted echo signal vector as in step 4. y ^{^} (j) is deducted,
The TF converter 431 obtains the residual signal vector E (j) in the frequency domain.

The coherence estimation unit 432 calculates the coherence from the residual signal vector E (j) in the frequency domain and the input signal vector X _m (j) in the frequency domain according to step 5. The specific configuration of the coherence estimation unit 432 is shown in FIGS. 8 and 9. The first and second correlation removing units 4321 _m , 43 corresponding to each frequency band
The residual signal vector E (j) and the input signal vector X _m (j) in the frequency domain are input to 22 _m , and the coherence calculation unit 4323 ₁
The coherence is calculated from ˜4323 _M , and the echo component ratio calculation unit 4324 obtains the ratio of the echo component in the residual signal.

The filter updating units 433 _{1 to} 433 _M obtain a correction vector in the frequency domain in accordance with step 6 from the input signal vector X _m (j) in the frequency domain and the residual signal vector E (j) in the frequency domain, and at the same time Compute the matrix P (j) according to 7. And step 8
The correction vector is corrected according to to update the adaptive filter coefficient. The updated filter coefficient is used by the filter processing unit 4
Passed to 12 _{1 to} 412 _M. Example 2 In Example 2, a step size control method based on coherence is described in the literature: Emura, Haneda, “Additional Signal Enhancement Type Frequency Domain Stereo Adaptive Algorithm”, Acoustical Society of Japan 200
1st Fall Research Presentation, pp. 537-538 (200
A case will be described in which the residual signal is applied to the multi-channel adaptive algorithm proposed in 1) and the residual signal is used as the target signal.

This adaptive algorithm uses the input signal x
determining a correction vector of the adaptive filter from the modified signal z _m (k) instead of _m (k). Therefore, in addition to the M channel received signal u _m (k), the M channel echo canceller 7 of FIG.
M-channel additional signal generated by the correlation variation processing section _{6 1 ~6 M g m (u} m (k)) is also input. The correlation fluctuation processing units 6 _{1 to} 6 _M are devices generally used for improving the echo path estimation performance of the multi-channel echo canceller. The M-channel echo canceller 7 in FIG. 4 updates the coefficient of the adaptive filter according to the following steps.

(Step 1) Received signal u of each channel
_From the additional signal g _m (u _m (k)) obtained by inputting _m (k) and the received signal u _m (k) to the correlation fluctuation processing unit 6 _M , the reproduced signal x
_m (k) and the correction signal z _m (k) are x _m (k) = u _m (k) + g _m (u _m (k)) z _m (k) = au _m (k) + g _m (u _m (k)) (where, m = 1, ..., M, 0 <a ≦ 1). And 2L length for each L / D sample
Of the signal vector of X _m (j) = diag (FFT ([x _m (jL / D-2L
+1), ..., X _m (jL / D)] ^T )) Z _m (j) = diag (FFT ([z _m (jL / D-2L
+1), ..., Z _m (jL / D)] ^T )) (where m = 1, ..., M).

(Step 2) X _m (j) in the frequency domain
And H ^{^} _m (j) are multiplied to filter the input signal vector for each channel. Inverse FFT processing is performed on the calculation result, and the time-domain signal vector y ^{^} _m (j) (where m
= 1, ..., M) is obtained. _{^{y ^ m (j) = [}} 0 L I L] IFFT (x
_m (j) H ^{^} _m (j)) where 0 _L is an L × L zero matrix and I _L is an L × L identity matrix. (Step 3) Signal vector y ^{^} _m (j) (m = 1,
, M), and the vector y of the predicted echo signal
^{^} (J) is obtained.

Y ^{^} (j) = Σ ^M _{m = 1} y ^{^} _m (j) (Step 4) Sound pickup signal vector y in the time domain
A residual signal vector is obtained from (j) and the predicted echo signal vector y ^{^} (j), and is transformed into the frequency domain by FFT.

[0059]

[Equation 25] (Step 5)

[0060]

[Equation 26]

[0061]

[Equation 27]

(Step 6) The residual signal and the correction signal are processed in the frequency domain to obtain a correction vector dH ^{^} _m (j).

[0063]

[Equation 28]

Calculate by However, the function X _m (j,
f) and Z _m (j, f) are the matrix X _m (j) and the matrix Z
_{It is} the (f, f) th element of _m (j). δ has a denominator of 0
It is a small positive constant for preventing P (j, f) in the matrix P (j) is the sum of the cross spectra of the input signal of each channel and the correction signal. (Step 8) From the coherence γ ² (f) obtained in Step 5

[0065]

[Equation 29]

A matrix M (j) having diagonal elements of coherence γ ² (f) is obtained by However, μ ₀ is set to a fixed value between 0 and 1. Update the adaptive filter with the following formula. H ^{^} _m (j + 1) = H ^{^} _m (j) + M (j) P
(J) dH ^{^} _m (j) By multiplying the matrix M (j), the step size is optimally controlled based on the ratio of the reverberant component to the target signal for each frequency band. Matrix P (j) is modified vector d
It is known that applying H ^{^} _m (j) corresponds to whitening processing of an input signal, and improves the convergence characteristic of an adaptive filter when the input signal is a chromatic signal such as voice.

The inside of the M-channel echo canceller 7 has a structure as shown in FIG. Reproduction signal x _m (k) and correction signal z _m
TF conversion units 702 _m and 705 _m that perform TF conversion on (k) correspond to the TF conversion unit 411 _m in FIG. 3. Adder 701 _m
Additional signal to the received signal _{u m (k) g m (} u m (k)) is summed reproduced signal x _m (k) is generated by, TF conversion unit 702 _m
Is converted into a matrix X _m (j) by. The received signal u _m (k) is multiplied by a by the attenuator 703 _m (where a is a value from 0 to 1), and the additional signal g _m (u _m (k)) is added by the adder 704 _m. As a result, the correction signal z _m (k) is generated. Then, the TF conversion unit 705 _m causes the matrix Z
converted to _m (j).

The matrix X _m (j) is the filter processing unit 712 _m.
And the matrix Z _m (j) is passed to the filter updating unit 733 _m . Filter processing unit 712 _m , FT conversion unit 71
3 _m , the vector addition unit 714 generates the predicted echo signal through the processes of step 2 and step 3. The sound pickup signal y (k) obtained from the microphone 3 is divided into blocks by the blocking unit 75, and the difference from the predicted echo signal vector is calculated by the vector subtraction unit 72 according to step 4, and T
The F conversion unit 731 converts the frequency domain. The coherence estimation unit 732 determines the residual signal vector E in the frequency domain.
Step 5 from (j) and the input signal vector X _m (j)
Estimate the coherence according to. Filter updating unit 73
3 _m (m = 1, ..., M) corresponds to Step 6, Step 7,
Update H ^{^} _m (j) in the frequency domain according to step 8.

The numerical simulation result of the second embodiment will be described with reference to FIG. This numerical simulation
The number of input channels is M = 2 and the sampling frequency is 8
The echo was generated by setting the frequency to kHz and cutting off the room transfer function measured in a room with a reverberation time of 200 ms as an echo path to 700 taps. As the interfering signal, a signal in which the level-varying Phos noise and the transmission signal were superimposed was used. The echo signal, the disturbing signal, the picked-up signal = the echo signal + the disturbing signal, and the residual signal e (k) after applying this method are as shown in FIG. Using this signal, the conventional method without step size control and the proposed step size control method were compared.

Let L be the number of adaptive filter taps per channel.
= 512 and the adaptive filter has 128 samples or 16
D = 4 is set so that it is updated every ms. Also, μ ₀
= 0.2 was set. The change in the coefficient error of the adaptive filter is shown in FIG. According to this graph, the coefficient error due to the estimation is deteriorated in the conventional method (dotted line) in the section (t = 4 to 6 s) in which the interference signal is slightly larger. However, with the proposed method (solid line), the estimation of this section is stable. In addition, an interval (t = 6) in which the interfering signal rapidly increases
s), the conventional method has a coefficient error of 0 dB to 8 dB.
The echo path estimation becomes unstable due to the expansion to. on the other hand,
In the proposed method, the deterioration of the coefficient error in this section is -6 dB to -5 dB, which is 1 dB.

[0071]

As described above, according to the present invention, the product of the reciprocal of the conventional correction vector and the input signal power is used as the correction amount between the adaptive filter coefficient in the frequency domain and the filter coefficient in the immediately preceding frame. By correcting the residual signal or the picked-up signal and the coherence between the input signal, the echo path of the adaptive filter can be estimated even in the presence of interference signals other than echo, such as speech, ambient noise, etc. You can be robust.

[Brief description of drawings]

FIG. 1 is a diagram illustrating an outline of an entire multi-channel acoustic communication device.

FIG. 2 is a diagram illustrating a conventional example.

FIG. 3 is a diagram illustrating an example.

FIG. 4 is a diagram illustrating an outline of an entire multi-channel acoustic communication device including an embodiment.

FIG. 5 is a diagram illustrating another embodiment.

FIG. 6 is a diagram showing an echo signal, an interference signal, and a sound pickup signal.

FIG. 7 is a diagram showing a numerical simulation result of an example.

FIG. 8 is a diagram for explaining calculation of coherence and echo component ratio.

FIG. 9 is a diagram illustrating a correlation component removal calculation.

Claims (11)

[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, and M channel signals are reproduced from the speakers to correspond to each microphone. A multi-channel acoustic in which an echo signal is predicted by inputting an M channel reproduction signal to the adaptive filter, the predicted echo signal is subtracted from the sound pickup signal, and the adaptive filter coefficient is updated so as to reduce the obtained residual signal. In the communication system, the residual signal is used as a target signal, the ratio of the reverberation component in the target signal is calculated, and the adaptive filter coefficient is updated using this information. 2. 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, and M channel signals are reproduced from the speakers to support each microphone. A multi-channel acoustic in which an echo signal is predicted by inputting an M channel reproduction signal to the adaptive filter, the predicted echo signal is subtracted from the sound pickup signal, and the adaptive filter coefficient is updated so as to reduce the obtained residual signal. In the communication system, a sound pickup signal is used as a target signal, the ratio of the reverberation component in the target signal is calculated, and the adaptive filter coefficient is updated using this information. 3. The echo canceling method according to claim 1, wherein
(A) The M channel playback signal is converted into the frequency domain for each short time period, and the frequency domain adaptive filter coefficient is multiplied,
Transform the time domain to predict the echo signal, (B) subtract the predicted echo signal from the picked-up signal, transform the residual signal into the frequency domain for each short time interval, and (C) regenerate the signal. From the short-time spectrum of the target signal, find the ratio of the reverberation component in the target signal for each frequency band, and (D) modify the correction vector obtained by multiplying the residual signal and the playback signal for each frequency band in the frequency domain. An echo canceling method, which comprises the step of updating the adaptive filter coefficient by correcting for each frequency band based on the ratio of the echo component in the signal and the information of the input signal and the correction signal. 4. The echo canceling method according to claim 1, wherein
(A) M channel received signal is processed to generate an M channel additional signal whose inter-channel correlation can be regarded as almost uncorrelated,
(B) The M channel additional signal is added to each of the received signals to obtain a reproduction signal, and (C) the M channel reproduction signal is converted into the frequency domain for each short time interval, and the adaptive filter coefficient in the frequency domain is multiplied to obtain the time domain. (E) The residual signal between the picked-up signal and the predicted echo signal is converted into the frequency domain for each short time interval, and (E) the reproduction signal and the target signal From the spectrum, find the ratio of the reverberant component to the target signal for each frequency band, and (E)
A correction signal is generated by adding the M channel reception signal multiplied by a times (a is a value of 0 to 1) to the M channel addition signal, and (F) the correction signal is converted into the frequency domain for each short time interval,
(G) The correction vector obtained by multiplying the residual signal and the correction signal for each frequency band in the frequency domain is used for each frequency band based on the ratio of the reverberation component in the target signal and the information of the input signal and the correction signal. The echo canceling method, which comprises the step of: 5. The echo canceling method according to claim 1, 2, 3, or 4.
Be careful M-th channel reproduction signal x_m(K) (m = 2, ...,
M) to the reproduction signal x of the 1st to m-1st channels₁(K), ...
., X_m-1Short time of signal without correlation component with (k)
Spectrum [Equation 1] Seeking From the target signal v (k), the 1st to (m-1) th channel reproduction signals
x₁(K), ..., x _m-1Remove the correlation component with (k)
Short time spectrum of the signal [Equation 2] Seeking Two short time spectra [Equation 3] About coherence [Equation 4] Seeking Is the short-time spectrum of the 1st channel playback signal and the target signal?
Coherence γ² _1vFind (f), The ratio of the reverberant component to the target signal v (k) [Equation 5] An echo canceling method characterized by being obtained in. 6. Reverberation from an M channel reproduction signal for each microphone 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. An adaptive filter that predicts the signal and a means that obtains a residual signal by subtracting the predicted echo signal from the sound pickup signal by the microphone, and a means that obtains the ratio of the echo component in the target signal with the residual signal as the target signal and the target signal. Means for updating the adaptive filter coefficient so as to reduce the obtained residual signal by using the ratio of the echo component occupying in the echo canceller. 7. Reverberation from an M channel reproduction signal for each microphone, which is 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. An adaptive filter for predicting the signal, a means for subtracting the predicted echo signal from the sound pickup signal from the microphone to obtain a residual signal, a means for obtaining the residual signal as the target signal, and a means for obtaining the ratio of the reverberation component in the target signal, An echo canceller comprising means for updating the adaptive filter coefficient so as to reduce the residual signal by using the ratio of the echo component in the target signal. 8. The echo canceller according to claim 6, wherein the means for processing the M channel received signal and generating the M channel additional signal whose inter-channel correlation can be regarded as substantially uncorrelated and the M channel additional signal are received respectively. A means for adding the signal to the reproduced signal, a means for converting the M channel reproduced signal into the frequency domain for each short time interval, a means for multiplying the adaptive filter coefficient in the frequency domain, and a predicted echo signal for conversion into the time domain A means for obtaining the residual signal of the collected signal and the predicted echo signal into a frequency domain for each short time interval, a means for obtaining a short time spectrum of the reproduced signal and the target signal, and a reproduced signal A means for obtaining the ratio of the reverberation component occupying in the target signal for each frequency band from the short-time spectrum of the target signal, and an M channel reception signal in which the M channel additional signal is multiplied by a (a is a value of 0 to 1) A means for adding signals to generate a correction signal, a means for converting the correction signal into the frequency domain for each short time period, and a correction by multiplying the residual signal and the correction signal for each frequency band in the frequency domain A means for obtaining a vector, a means for correcting the correction vector for each frequency band based on the ratio of the reverberation component in the target signal and the information of the input signal and the correction signal, and an adaptive filter coefficient using the corrected correction vector. An echo canceller having means for updating. 9. The echo canceller according to claim 6, 7, or 8, further comprising: means for obtaining a short-time spectrum of a signal obtained by removing a correlation component between the m-th channel reproduced signal and the 1st to (m-1) th channel reproduced signals. , A means for obtaining a short-time spectrum of a signal from which a correlation component with the first to m-1th channel reproduction signals has been removed from the target signal, a means for obtaining coherence of two short-time spectra, and a first channel reproduction An echo canceller comprising: means for obtaining a coherence from a short-time spectrum of a signal and a target signal; and means for obtaining a ratio of an echo component occupying the target signal from the coherence. 10. An echo canceling program for executing the echo canceling method according to claim 1 by a computer. 11. A computer-readable recording medium in which the echo canceling program according to claim 10 is recorded.