JP5391103B2 - Multi-channel echo canceling method, multi-channel echo canceling apparatus, multi-channel echo canceling program and recording medium therefor - Google Patents

Description

  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) are arranged in a common sound field, and an M channel received signal is reproduced from the speakers and picked up by a microphone. The present invention relates to a technique for erasing acoustic echo (hereinafter simply referred to as “echo”) from a collected sound signal (hereinafter referred to as “first collected signal”), and more particularly to a technique for eliminating echo in a loudspeaker communication system such as a video conference system.

  The received signal is reproduced by the speaker, and the sound is picked up by the microphone to generate an echo. If it is transmitted as it is, problems such as troubles in communication and discomfort arise. Further, howling occurs when the volume of the speaker or microphone is high, making it impossible to make a call. In particular, such a problem becomes conspicuous in the voice call system.

  In order to solve this problem, there is an echo canceller that cancels echoes using an adaptive filter as a prior art. Non-Patent Document 1 is known as a conventional multi-channel echo canceller. A conventional multi-channel echo canceller 90 will be described with reference to FIG.

  M speakers 21,..., 2M and N microphones 31,..., 3N are arranged in a common sound field, and the M channel reception signals x1 (k),. Is reproduced, the multi-channel echo canceller 90 cancels the reproduced sound (echo) that wraps around the microphone via m × n echo paths hmn (k). However, m = 1,..., M, n = 1,..., N, M is 2 or more, and N is 1 or more. The multi-channel echo canceller 90 is connected to all M channel receiving terminals 11 to 1M on the receiving side, all N channel transmitting terminals 41,..., 4N and microphones 31,. The received signals x1 (k),..., XM (k) and the collected sound signals y1 (k),..., YN (k) are input, and the transmitted signals u1 (k),. 41,..., 4N. The multi-channel echo canceller 90 includes N echo cancelers 801,..., 80N, and each echo canceler includes an echo predictor 81, a subtractor 82, and an echo path estimator 83. FIG. 1 illustrates the echo canceller 801, where y1 (k) is y (k), u1 (k) is u (k), h11 (k),..., HM1 (k) is h1 (k), ..., expressed as hM (k). The same processing can be performed on the collected sound signals from other microphones, and it is only necessary to arrange the configurations in FIG. 1 in parallel.

  The multi-channel echo canceller 90 obtains a difference between the collected sound signal y (k) and the pseudo echo signal y ′ (k), that is, an error signal u (k) by the subtracting unit 82, and receives this signal u (k) and the received signal. The echo path estimation unit 83 sequentially estimates the echo path (filter coefficient h ′ (k)) from the signals x1 (k),..., XM (k), and the echo prediction unit 81 uses the estimation result to estimate the pseudo echo signal. y ′ (k) is generated. In a state where the echo path estimation is performed with high accuracy, the echo component included in the collected sound signal y (k) and the pseudo echo signal y ′ (k) are almost equal, and the error signal u (k) includes almost no echo. It will not be.

M.M.Sondhi, D.R.Morgan, and J.L.Hall, “Stereophonic Acoustic Echo Cancellation-An Overview of the Fundamental Problem”, IEEE Signal Processing Letters, AUGUST 1995, vol.2, no.8, pp.148-151

  However, the prior art cannot always sufficiently cancel the echo.

  Since the echo path estimation by the echo path estimation unit 83 is not completed instantaneously, the residual echo increases every time the echo path changes due to human movement or the like.

  In the double talk state, since the error signal includes the voice of the sender, the estimation of the echo path is disturbed and the residual echo becomes large.

  Further, as described in Non-Patent Document 1, when the received signal is multi-channel, since the correlation between channels is high, the estimated echo path and the true path are true even when the echo is canceled. The echo paths may not always match. In that case, when the speaker on the receiving terminal side changes and the cross-correlation between channels of the received signal changes, the residual echo suddenly increases.

  Then, there is a problem that the call quality deteriorates when the residual echo becomes large.

  In order to solve the above-described problem, the multi-channel echo cancellation technique according to the present invention has arranged M speakers (M is an integer of 2 or more) and N microphones (N is an integer of 1 or more) in a common sound field. When the M channel received signal is reproduced from the speaker, an echo that goes around the microphone via the echo path is deleted, and a signal (hereinafter, referred to as a signal obtained by using the M channel received signal and the first sound pickup signal collected by the microphone) (Referred to as “sound pickup signal etc.”) is converted into a frequency domain signal, and the power spectrum Pmm (f, j) of each channel of the M channel received signal and the cross spectrum Pm′m ( f, j) and the cross spectrum Q (f, j) between the M channel received signal and the collected sound signal, etc., are obtained, and the power spectrum Pmm (f, j), the cross spectrum Pm′m ( , J) and Q (f, j) are used to estimate the input / output transfer characteristics for each frequency, and the echo component contained in the collected sound signal or the like is determined from the M-channel received signal in the frequency domain and the estimated input / output transfer characteristics. Predict and subtract the predicted echo component from the frequency domain collected sound signal, etc. to obtain the transmitted signal, convert the transmitted signal obtained in the subtraction step into a time domain signal, and collect the collected sound signal with a microphone. It is either the first collected sound signal or the second collected signal obtained by subtracting the pseudo echo signal from the first collected signal.

  The present invention relates to the power spectrum Pmm (f, j) of each channel of the M channel received signal, the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, the M channel received signal and the sound collection. In order to estimate the input / output transfer characteristics from the cross spectrum Q (f, j) between the signals, etc., and to quickly predict the echo component contained in the collected sound signal from the M channel received signal and the input / output transfer characteristics, There is an effect that the echo can be quickly canceled without being affected by the cross correlation between the channels of the fluctuation, the double talk and the received signal.

The figure which shows the structural example of the conventional multichannel echo cancellation apparatus. 1 is a diagram illustrating a configuration example of a multi-channel echo canceling apparatus according to Embodiment 1. FIG. FIG. 3 is a diagram illustrating a processing flow of the multi-channel echo canceling apparatus according to the first embodiment. The figure which shows the structural example of an input-output correlation coefficient calculation part. FIG. 5 is a diagram illustrating a configuration example of a multi-channel echo canceling apparatus according to a second embodiment. FIG. 6 is a diagram illustrating a processing flow of the multi-channel echo cancellers according to the second and third embodiments. FIG. 10 is a diagram illustrating a configuration example of a multi-channel echo canceling apparatus according to a third embodiment. The figure which shows the structural example of a transfer characteristic adjustment part.

  Hereinafter, embodiments of the present invention will be described in detail.

<Multi-channel echo canceller 100>
A multi-channel echo canceling apparatus 100 according to the first embodiment will be described with reference to FIGS. The multi-channel echo canceling apparatus 100 includes N residual echo canceling units 6001,..., 600N, and each residual echo canceling unit includes M frequency domain transforming units 6011,..., 601M, and a frequency domain transforming unit 602. , An input / output correlation coefficient calculation unit 603, an input / output transfer characteristic estimation unit 604, a residual echo prediction unit 605, a subtraction unit 606, and a time domain conversion unit 607.

  M speakers 21,..., 2M and N microphones 31,..., 3N are arranged in a common sound field, and the M channel reception signals x1 (k),. Is reproduced, the multi-channel echo canceling apparatus 100 cancels the reproduced sound (echo) that wraps around the microphone via m × n echo paths hmn (k). However, m = 1,..., M, n = 1,..., N, M is 2 or more, and N is 1 or more. The multi-channel echo canceling apparatus 100 is connected to all M channel receiving terminals 11 to 1M on the receiving side, all N channel transmitting terminals 41,..., 4N and microphones 31,. The received signals x1 (k),..., XM (k) and the collected sound signals y1 (k),..., YN (k) are input, and the transmitted signals v1 (k),. 41,..., 4N.

  FIG. 2 illustrates the residual echo canceller 6001, where y1 (k) is y (k), v1 (k) is v (k), and h11 (k),..., HM1 (k) is h1 (k). ,..., HM (k). Similar processing can be performed on the collected sound signals from other microphones, and the configuration of FIG. 2 only needs to be arranged in parallel.

<Frequency domain transforming units 6011,..., 601M and 602>
.., 601M receive M channel received signals x1 (k),..., XM (k) as inputs, respectively, and frequency domain signals X1 (f, j),..., XM (f, j) And output (s6011,..., S601M). However, f represents a frequency number and j represents a frame number.

  The frequency domain conversion unit 602 receives the first collected sound signal y (k) collected by the microphone, converts it to a frequency domain signal Y (f, j), and outputs the signal (s602).

  A case will be described in which each signal is set to 1 frame = 2 L samples, the L / D samples are blocked, and the L / D samples are shifted to create a frame. However, L is a natural number, D is a natural number that divides L, and time k = jL / D. The conversion to the frequency domain is performed by, for example, FFT (Fast Fourier transform) or DFT (discrete Fourier transform), and L may be a power of 2 in order to simplify and speed up the calculation. For example, L = 64 to 512, D = 2 to 8 and the like.

<Input / output correlation coefficient calculation unit 603>
The input / output correlation coefficient calculation unit 603 receives the M channel reception signal X1 (f, j),..., XM (f, j) and the first sound pickup signal Y (f, j) as inputs, and receives the M channel reception signal. X1 (f, j),..., XM (f, j), the power spectrum Pmm (f, j) of each channel, and the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, A cross spectrum Q (f, j) between the M channel received signal and the collected sound signal is obtained and output (s603). In other words, the input / output correlation coefficient calculation unit 603 inputs the correlation coefficient P (f, j) between the input signals composed of the power spectrum Pmm (f, j) and the cross spectrum Pm′m (f, j) and the input signal. A correlation coefficient Q (f, j) between output signals is obtained.

The input / output correlation coefficient calculation unit 603 will be described with reference to FIG. For example, the input / output correlation coefficient calculation unit 603 includes a power spectrum calculation unit 603a, a received signal cross spectrum calculation unit 603b, and a received / acquired sound signal cross spectrum calculation unit 603c.

  The power spectrum calculation unit 603a calculates the power spectrum Pmm (f, j) of each channel of the received signal using Xm (f, j).

  The inter-received signal cross spectrum calculation unit 603b calculates a cross spectrum Pm′m (f, j) between each channel of the received signal using X1 (f, j),..., XM (f, j). However, m ′ ≠ m.

  The received-acquired sound pickup signal cross spectrum calculation unit 603c has a cross spectrum Q (f, j) between X1 (f, j),..., XM (f, j) and the first acquired sound signal Y (f, j). Is calculated.

  For example, Pmm (f, j), Pm′m (f, j), and Q (f, j) are the M channel received signal Xm (f, j) and the first collected sound signal Y at time k = jL / D. From (f, j)

Calculated by X ^* means a complex conjugate of X, E [] means taking an average, and as an example of the averaging process,

As described above, there are a method using a processing result of one frame before and a smoothing constant β that takes a value of 0 to 1, a method of obtaining by multiplying a past several frames by a time constant, and the like. The same applies to Q (f, j).

  When m = m ′, the power spectrum Pmm (f, j) of each channel is obtained, and when m ≠ m ′, the cross spectrum Pm′m (f, j) between the channels of the received signal is obtained. 1 ≦ m ≦ M and 1 ≦ m ′ ≦ M.

<Input / output transfer characteristic estimation unit 604>
The input / output transfer characteristic estimation unit 604 receives the power spectrum Pmm (f, j), the cross spectrum Pm′m (f, j), and Q (f, j) as input, and uses these to input / output transfer for each frequency. The characteristic G (f, j) is estimated and output (s604).

  For example, the input / output transfer characteristic estimation unit 604 calculates the input / output transfer characteristic G (f, j).

Estimated by

  For the matrix composed of the power spectrum and the cross spectrum, a small constant may be added to the diagonal component in order to stabilize the inverse matrix calculation.

<Residual echo prediction unit 605>
The residual echo prediction unit 605 receives the input / output transfer characteristics G (f, j) estimated as M-channel received signals X1 (f, j),..., XM (f, j) in the frequency domain, and uses them. An echo component included in the first collected sound signal Y (f, j) is predicted and output (s605).

  For example, the echo component

To predict.

<Subtraction unit 606 and time domain conversion unit 607>
The subtracting unit 606 receives the first sound pickup signal Y (f, j) and the predicted echo component Y ^ (f, j) as input, and predicts from the first sound pickup signal Y (f, j) in the frequency domain. The echo component Y ^ (f, j) is subtracted to obtain and output the transmission signal V (f, j) (s606).

V (f, j) = Y (f, j) -Y ^ (f, j) (8)
The time domain conversion unit 607 receives the transmission signal V (f, j) obtained by the subtraction unit 606, converts it into a time domain signal v (k), and outputs it to the transmission terminal 41 as a transmission signal. (S607).

<Effect>
Between the power spectrum Pmm (f, j) of each channel of the M channel received signal, the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, and between the M channel received signal and the collected sound signal, etc. It is necessary to update the filter coefficient of the adaptive filter in order to estimate the input / output transfer characteristics from the cross spectrum Q (f, j) and predict the echo component included in the collected sound signal from the M channel received signal and the input / output transfer characteristics. There is no. Therefore, it is possible to respond immediately to fluctuations in the echo path, predict an echo faster than conventional multi-channel echo cancellers, and suppress estimated fluctuations due to signals other than echoes. In addition, since the amplitude and phase of the transfer characteristics and echoes are estimated in the frequency domain and the echo is canceled by subtraction, the distortion of transmission loss can be reduced even during double talk, and the echo can be canceled regardless of the number of channels. it can.

<Modification>
Since the values of M-channel received signals X1 (f, j) to XM (f, j) in the frequency domain used in each residual echo canceller are 6001 to 600N, N residual echo cancelers 6001 to 600N The output of the frequency domain transform unit of the received signal may be shared.

  Even if echo cancellation processing is not performed for all transmission signals, echoes may be canceled only for some transmission signals.

  The input to the multi-channel echo canceller 100 may be a received signal obtained by performing a correlation variation process on the signal obtained from the receiving terminals 11,.

  The above-described multi-channel echo canceller can also be operated by a computer. In this case, each process of the program for causing the computer to function as the target device (the device having the functional configuration shown in the drawings in the embodiment) or the processing procedure (shown in the embodiment) is executed on the computer. The program may be downloaded into a computer from a recording medium such as a CD-ROM, a magnetic disk, or a semiconductor storage device or via a communication line, and the program may be executed.

<Multi-channel echo canceller 200>
A multi-channel echo canceling apparatus 200 according to the second embodiment will be described with reference to FIGS. In FIG. 5, the processing of the collected sound signal from the microphone 31 will be described. However, the same processing can be performed on the collected sound signal from other microphones, and the configuration of FIG. 5 only needs to be arranged in parallel. . The multi-channel echo canceller 200 has N echo cancelers 801,..., 80N and N residual echo cancelers 6001,. The residual echo canceling unit 6001 has the same configuration as the residual echo canceling unit 6001 of the multi-channel echo canceling apparatus 100 according to the first embodiment, and instead of the first sound collecting signal y (k), the second sound collecting signal u ( The difference is that k) is input. The second collected sound signal u (k) is an error signal that is an output of the echo canceling unit 801. Therefore, the residual echo canceling unit 6001 replaces the echo component included in the first sound pickup signal y (k) picked up by the microphone, and the second sound pickup signal obtained by subtracting the pseudo echo signal from the first sound pickup signal. (Error signal) The residual echo component included in u (k) is eliminated.

  For example, the frequency domain conversion unit 602 receives the error signal output from the echo cancellation unit 801 as the second collected sound signal u (k), converts the error signal into a frequency domain signal U (f, j), and outputs the signal. .

  The input / output correlation coefficient calculation unit 603 receives the M channel received signal X1 (f, j),..., XM (f, j) and the second collected sound signal U (f, j) as inputs, and receives the M channel received signal. X1 (f, j),..., XM (f, j), the power spectrum Pmm (f, j) of each channel, and the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, A cross spectrum Q (f, j) between the M channel reception signal and the second sound pickup signal is obtained and output.

The residual echo prediction unit 605 receives the input / output transfer characteristics G (f, j) estimated as M-channel received signals X1 (f, j),..., XM (f, j) in the frequency domain, and uses them. An echo component included in the second collected sound signal U (f, j) is predicted and output.

To predict.

  The subtraction unit 606 receives the second collected sound signal U (f, j) and the predicted echo component U ^ (f, j) as input, and predicts from the second collected sound signal Y (f, j) in the frequency domain. The echo component U ^ (f, j) is subtracted to obtain and output the transmission signal V (f, j).

V (f, j) = U (f, j) -U ^ (f, j) (8 ')
The first sound collection signal Y (f, j) or the second sound collection signal U (f, j) corresponds to the sound collection signal etc. S (f, j) in the claims, and is predicted by the residual echo prediction unit 605. The echo component Y ^ (f, j) included in the first collected sound signal or the echo component U ^ (f, j) included in the second collected sound signal is the echo component included in the collected sound signal etc. It corresponds to S ^ (f, j).

<Echo elimination unit 801>
On the other hand, the echo cancellers 801,..., 80N have the same configuration as the conventional echo canceller 90. For example, the echo canceling unit 801 obtains an error signal as follows. The echo cancellation unit 801 includes an echo prediction unit 81, a subtraction unit 82, and an echo path estimation unit 83.

  First, assuming that the impulse response of the echo path from the speaker 2m to the microphone 31 is hm (k) and the length thereof is L, the following relationship exists between the M channel received signal and the collected sound signal y (k).

Impulse response and received signal of each channel
hm = [hm (0)… hm (L-1)] ^T (12)
xm = [xm (0)… xm (L-1)] ^T (13)
As a vector, the relationship between the M channel received signal and the first collected sound signal is described as follows.

y (k) = h1 ^T x1 (k) +… + hM ^T xM (k) (14)
However, T represents transposition.

<Echo Prediction Unit 81>
The echo prediction unit 81 inputs the received signals x1 (k),..., XM (k) to the pseudo echo path by the adaptive filter, and generates and outputs a pseudo echo signal y ′ (k) (s81). The echo prediction unit 81 includes an adaptive filter, and the characteristic of the adaptive filter is controlled by an echo path estimation unit 83 (to be described later) so that the error signal of the subtraction unit 82 in the reception state is minimized.

For example, the filter coefficient of the adaptive filter for each channel
h'm = [h'm (0)… h'm (L-1)] ^T (15)
age,
y '(k) = h'1 ^T x1 (k) +… + h'M ^T xM (k) (16)
Is generated.

<Subtraction unit 82>
The subtracting unit 82 receives the first collected sound signal y (k) and the pseudo echo signal y ′ (k), subtracts the pseudo echo signal y ′ (k) from the first collected sound signal y (k), The collected sound signal u (k) is obtained and output (s82).

u (k) = y (k) -y '(k) (17)
The obtained second collected sound signal (error signal) u (k) is output to the echo path estimation unit 83 and the frequency domain conversion unit 602 in the residual echo cancellation unit 6001.

<Echo path estimation unit 83>
The echo path estimation unit receives the second collected sound signal u (k) and the received signals x1 (k),..., XM (k) and updates the filter coefficient h ′ (k) of the adaptive filter using them. And output (s83). A case where the Normalized Least Mean Square algorithm (NLMS algorithm) is used as a coefficient correction method for the adaptive filter will be described.

h'm (k + 1) = h'm (k) + μu (k) xm (k) (18)
Where μ is the step size,

Determined by. Control is based on μ0 and the power of the input signal. However, μ0 is a parameter set in advance to a value of 0 to 1 in order to stabilize the estimation. The echo path estimation unit 83 copies the updated filter coefficient h ′ (k + 1) and outputs it to the echo prediction unit 81.

<Effect>
By adopting such a configuration, it is possible to obtain the same effects as those of the first embodiment while maintaining the echo cancellation amount and the voice quality equivalent to those of the prior art.

  Residual echoes can be suppressed by performing echo cancellation processing by the high-speed residual echo canceling unit in the case where the initial stage or situation where it takes time to estimate the echo path in the echo canceling unit 80n changes. Further, when the echo path estimation in the echo canceling unit is stable, the residual echo can be further suppressed by the performance of the echo canceling unit. Therefore, compared with the case where the echo canceling unit and the residual echo canceling unit are used alone, it is possible to perform a call with the residual echo reduced over the entire processing time.

<Modification>
The filter coefficient may be updated by a conventional technique other than the learning identification method (for example, a projection algorithm, an exponential weighting algorithm, an exponential weighted projection algorithm, etc.).

  Similar to the first embodiment, the residual echo cancellers 6001 to 600N may share the output of the frequency domain transform unit of the received signal.

<Multi-channel echo canceller 300>
A multi-channel echo canceling apparatus 300 according to the third embodiment will be described with reference to FIGS.

  The multi-channel echo canceller 300 is different from the multi-channel echo canceller 200 in that it includes a transfer characteristic adjustment unit 3604.

<Transfer characteristic adjustment unit 3604>
The transfer characteristic adjustment unit 3604 includes a size calculation unit 3604a, a determination unit 3604b, and an adjustment unit 3604c. When the estimated input / output transfer characteristic G (f, j) is larger than the reference value, the input / output is performed. A process for adjusting the size of the transfer characteristic G (f, j) to match the reference value, so-called clip process, is performed.

  The transfer characteristic adjustment unit 3604 receives the input / output transfer characteristic G (f, j), which is the output of the input / output transfer characteristic estimation unit 604, stores the input / output transfer characteristic G '(f, j). ) Or the input / output transfer characteristic G (f, j) as an input value is output as it is. Hereinafter, processing of each unit will be described.

  The magnitude calculator 3604a calculates, for example, a norm | G (f, j) | as the magnitude of the estimated input / output transfer characteristic G (f, j). For example,

Asking.

  The determination unit 3604b determines whether or not the norm | G (f, j) | is larger than the reference value C designated in advance. When the norm | G (f, j) | is smaller than the reference value C, the input / output transfer characteristic G (f, j) as an input value is output as it is. When the norm is larger than the reference value, an adjustment instruction is transmitted to the adjustment unit 3604c. The reference value (value to be clipped) C is set as, for example, [an assumed acoustic coupling amount of one acoustic path] × [number of acoustic paths (sqrt)].

  When the adjustment unit 3604c receives the adjustment instruction, the adjustment unit 3604c multiplies the input / output transfer characteristic G (f, j) by the reference value C, and further divides by the norm | G (f, j) | f, j) is adjusted (s3604). For example, it is represented by the following formula.

  The residual echo prediction unit 605 is configured to estimate the input / output transfer characteristic G (f, j) or the adjusted input / output transfer characteristic G as the M-channel received signal X1 (f, j),..., XM (f, j) in the frequency domain. A residual echo component included in the second collected sound signal U (f, j) is predicted from '(f, j).

<Effect>
The same effect as in the second embodiment can be obtained, and the input / output transfer characteristics are sometimes overestimated due to the correlation between the transmission and the reception by correcting the input / output transfer characteristics by comparing the threshold values in this way. Can be prevented.

  Usually, there is no correlation between the voice spoken by the sender and the received signal. However, if the frame of the processing unit is shortened, a false correlation may occur between the voice spoken by the sender and the received signal. In that case, the input / output transfer characteristics are overestimated by the equations (4) and (6), the value of Y ^ (f, j) in the equation (7) becomes larger than necessary, and the voice quality is deteriorated. there is a possibility. In this embodiment, the deterioration can be prevented.

<Modification>
The multi-channel echo canceller 100 according to the first embodiment and the transfer characteristic adjustment unit 3604 may be combined. Even in this case, it is possible to prevent the input / output transfer characteristic from being overestimated.

  Further, the magnitude of the input / output transfer characteristic may be obtained by a conventional technique other than Expression (21).

90, 100, 200, 300 Multi-channel echo cancellers 6001,..., 600N, 36001,... 3600N Residual echo cancellers 801..., 801N Echo cancellers 6011, ..., 601M, 602 Frequency domain converter 603 Input / output phase relationship Number calculation unit 604 Input / output transfer characteristic estimation unit 605 Residual echo prediction unit 606 Subtraction unit 607 Time domain conversion unit

Claims (6)

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. When an M channel received signal is reproduced from the speakers, the microphone is connected via an echo path. A multi-channel echo cancellation method for canceling echoes that wrap around
An echo prediction step of generating the pseudo echo signal by inputting the received signal to the pseudo echo path by the adaptive filter;
A subtraction step of subtracting the pseudo echo signal from a first sound pickup signal picked up by the microphone to obtain a second sound pickup signal;
An echo path estimating step of updating a filter coefficient of the adaptive filter using the second collected sound signal and the received signal ;
A frequency domain conversion step of converting the M channel received signal and the second sound pickup signal into a frequency domain signal;
The power spectrum Pmm (f, j) of each channel of the M channel received signal, the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, the M channel received signal and the second collected sound signal An input / output correlation coefficient calculating step for obtaining a cross spectrum Q (f, j) between the respective frequencies;
An input / output transfer characteristic estimation step for estimating an input / output transfer characteristic for each frequency using the power spectrum Pmm (f, j), cross spectrum Pm′m (f, j) and Q (f, j);
A residual echo prediction step for predicting, for each frequency, an echo component included in the second collected signal from the frequency domain M-channel received signal and the estimated input / output transfer characteristics;
A subtracting step of subtracting the predicted echo component from the second collected signal in the frequency domain to obtain a transmission signal for each frequency;
A time domain conversion step of converting the transmission signal obtained in the subtraction step into a time domain signal,
And a multi-channel echo cancellation method. A multi-channel echo canceling method of claim 1 Symbol placement,
A magnitude calculating step for calculating the magnitude of the estimated input / output transfer characteristic;
A determination step of determining whether or not the size of the input / output transfer characteristic is larger than a reference value designated in advance;
When the magnitude of the input / output transfer characteristic is larger than the reference value, the input / output transfer characteristic is multiplied by the reference value, and further divided by the magnitude of the input / output transfer characteristic. An adjustment step for adjusting the output transfer characteristic;
And a multi-channel echo cancellation method. The multi-channel echo cancellation method according to claim 1 or 2 ,
In the input / output correlation coefficient calculating step, the Pmm (f, j), Pm′m (f, j), and Q (f, j) are the received signal Xm (f, j) and the second collected sound signal. Let S (f, j) be

In the input / output transfer characteristic estimation step, the input / output transfer characteristic G (f, j) is

In the residual echo prediction step, the echo component is

Predict as
And a multi-channel echo cancellation method. 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. When an M channel received signal is reproduced from the speakers, the microphone is connected via an echo path. A multi-channel echo canceller that cancels echoes that wrap around
An echo prediction unit that generates the pseudo echo signal by inputting the received signal into the pseudo echo path by the adaptive filter;
A subtractor for subtracting the pseudo echo signal from a first sound pickup signal picked up by the microphone to obtain a second sound pickup signal;
An echo path estimator that updates a filter coefficient of the adaptive filter using the second collected sound signal and the received signal;
A frequency domain converter that converts the M-channel received signal and the second collected sound signal into a frequency domain signal;
The power spectrum Pmm (f, j) of each channel of the M channel received signal, the cross spectrum Pm′m (f, j) between the channels of the M channel received signal, the M channel received signal and the second collected sound signal An input / output correlation coefficient calculation unit that obtains a cross spectrum Q (f, j) between each frequency ,
An input / output transfer characteristic estimator that estimates input / output transfer characteristics for each frequency using the power spectrum Pmm (f, j), cross spectrum Pm′m (f, j), and Q (f, j);
A residual echo prediction unit that predicts, for each frequency, an echo component included in the second collected signal from the input / output transfer characteristics estimated as the M-channel received signal in the frequency domain;
A subtracting unit that subtracts the predicted echo component from the second collected sound signal in the frequency domain and obtains a transmission signal for each frequency ;
To have a, and the time domain converter for converting the transmission signal obtained by the subtraction unit into a signal in the time domain,
And a multi-channel echo canceller. The program for making a computer perform the method in any one of Claim 1 to 3 . A computer-readable recording medium on which the program according to claim 5 is recorded.

Images