JP2005323084A - Method, device, and program for acoustic echo-canceling - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To prevent the occurrence of howling, when a loudspeaker is changed in a multichannel acoustic echo-canceling device. <P>SOLUTION: Multichannel acoustic echoes are canceled by the combination of an adaptive loss control means and a multichannel acoustic echo canceling means in the acoustic echo canceling method. Echo path estimation error information is extracted, by including a noncausal component in a multichannel adaptive filter. An attenuation amount on the reception side or the transmission side in the adaptive loss control means is controlled by the extracted echo path estimation error information. The attenuation amount is maintained stably with respect to transitional fluctuations, such as switching of speakers. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an acoustic echo erasing method, an acoustic echo erasing apparatus, and an acoustic echo erasing program for erasing an acoustic echo that becomes an obstacle to a call in a loudspeaker communication system and sometimes causes howling.

FIG. 1 shows the structure of the voice call system. The voice uttered by the speaker at the point B is picked up by the microphone 2B, and is reproduced by the speaker 1A at the point A via the transmission path 3. Since this loudspeaker call system does not require a handset to be held in the hand like a conventional telephone call, it is possible to make a call while working and to have a natural conversation call. Therefore, it is widely used for communication conferences, videophones, loudspeakers and the like.
In this voice call system, acoustic echo is a problem. Sound reproduced from the speaker 1A in FIG. 1, is collected by the microphone 2A via an echo path h _A as indicated by an arrow, through the transmission path 4, is reproduced by the speaker 1B point B. This is an acoustic echo, and the speaker at point B hears that his / her voice is reproduced from the speaker 1B after transmission delay. This acoustic echo causes adverse effects such as call failure and discomfort in a loudspeaker call system. Further, when the closed loop gain of the signal formed at the points A and B is larger than 1, a howling phenomenon occurs and the call cannot be made.

In order to overcome such problems of the voice call system, the adaptive loss control unit 5 and the acoustic echo canceling unit 6 may be combined as shown in FIG. 2 (Patent Document 1).
The operation of the adaptive loss control unit 5 will be described with reference to FIG. The adaptive loss control unit 5 includes a reception side attenuator 31, a transmission side attenuator 32, and a loss control unit 33. The loss control unit 33 determines the transmission / reception state from the received signal x_ (k) from the transmission path 3 and the output signal z (k) from the acoustic echo canceling unit 6 (k is a parameter indicating time). In the transmission state, the reception signal is attenuated by the reception side attenuator 31, and in the reception state, the signal attenuated by the transmission side attenuator 32 is used as a transmission signal, so that the open loop gain between the reception signal and the transmission signal is increased. Keep it above 1 to prevent howling.

  The transmission state and the reception state are determined, for example, by comparing the magnitudes of the short-time powers Px_ (k) and Pz (k) of the reception signal x_ (k) and the output signal z (k) of the acoustic echo canceller 6. it can. The short-time power of the speaker reproduction signal x (k) is equal to or smaller than the short-time power of the reception signal x_ (k) by the reception side attenuator 31. In a normal environment, it can be assumed that the echo signal z (k) after passing through the echo path and the acoustic echo canceller 6 is smaller than the speaker reproduction signal x (k). , Px_ (k)> Pz (k) is established, and at this time, the loss control unit 33 determines that it is in the receiving state. If there is a transmission signal of a certain level or higher, Px_ (k) <Pz (k) is established, and at this time, the loss control unit 33 determines that the transmission state.

  The amount of attenuation applied to the reception side attenuator 31 or the transmission side attenuator 32 is controlled according to the state of the acoustic echo canceller 6. The acoustic echo cancellation unit 6 includes a pseudo echo signal generation unit 41, a subtracter 42, and an echo path estimation unit 43. The pseudo echo signal generation unit 41 is configured by an adaptive filter, and the echo path estimation unit 43 detects the fluctuation of the echo path according to the error signal of the subtractor 42 in the reception state so that the error signal is minimized by the detection signal. The characteristics of the adaptive filter are controlled. Therefore, hereinafter, the pseudo echo signal generation unit 41 may be simply referred to as an adaptive filter.

The state of acoustic echo cancellation by the adaptive filter can be examined from the magnitude of the acoustic coupling amount G from the speaker 1 to the acoustic echo cancellation unit 6. The acoustic coupling amount G uses the reception signal x (k) in the reception state and the short-time powers Px (k) and Pz (k) of the residual signal z (k),
G = Pz (k) / Px (k)
Can be estimated. For example, when it is desired to keep the open loop gain between the received signal and the remaining echo (residual signal) to be −20 dB or less, acoustic echo cancellation by the adaptive filter has not progressed, and the acoustic coupling amount G is −20 dB. When the value exceeds the attenuation amount, the attenuation amount may be set so that the product of the acoustic coupling amount G and the attenuation amount becomes −20 dB.

  In the acoustic echo cancellation unit 6, the echo path estimation unit 43 estimates an impulse response (acoustic echo path transmission characteristic) of the acoustic echo path, and the pseudo echo signal generation unit 41 estimates the estimated impulse response w (k) and the reception side attenuation. The reception signal x (k) passed through the device 31 is convolved to generate a pseudo echo signal y ^ (k), and the subtractor 42 subtracts the pseudo echo signal y ^ (k) from the output signal of the microphone 2. If the impulse response of the echo path is estimated well, the echo signal and the pseudo echo signal are almost equal, and the echo signal included in the microphone output signal is almost eliminated.

である。またμ（０＜μ＜２）はエコー経路推定速度の調整にもちいられるステップサイズである。 The echo path estimator 43 estimates the impulse response of the acoustic echo path using an adaptive algorithm. This estimation is performed when it is regarded as a reception state. When the NLMS algorithm is used as the adaptive algorithm, the impulse response is estimated as follows using the received signal x (k) and the error signal e (k) so that the power of the error signal e (k) is minimized.

It is. Μ (0 <μ <2) is a step size used for adjusting the echo path estimation speed.

In the above method, attenuation can be controlled according to the estimated state of the echo path. The echo path immediately after startup is not estimated and the residual echo is large. For this reason, the acoustic coupling amount G becomes large, the attenuation amount is set large, and the loss control in the adaptive loss control unit 5 mainly prevents howling. As the estimation of the echo path proceeds, the acoustic coupling amount G decreases, and the set attenuation amount decreases. When the estimation of the echo path is almost completed, the acoustic echo canceling unit 6 mainly prevents howling and echo. In the state where the echo is almost canceled by the monaural echo canceller, it can be considered that the impulse responses of the echo path and the pseudo echo path are substantially the same.
JP 7-226699 A

When the adaptive loss control and acoustic echo cancellation of the monaural communication conference system are applied to the multi-channel communication conference as shown in FIG. 3, the problem is that the attenuation on the receiving side or transmitting side cannot be obtained well.
In multichannel communication conferencing systems, the cross-correlation between channels of received signals is usually very high, and even if the echo is almost cancelled, the echo transfer characteristics estimated by the multichannel adaptive filter and the true The echo transfer characteristics are not always the same. In this regard (non-patent literature MM Sondhi, DR Morgan, and JL Hall, “Stereo-phonic Acoustic Echo Cancellation -An Overview of the Fundamental Problem,” IEEE Signal Processing Letters, Vol. 2, no. 8, pp. 148- 151 (1995)). Therefore, unlike the conventional method, it is impossible to know the echo path estimation status of the adaptive filter from the speaker reproduction signal and the output signal of the acoustic echo canceller via the acoustic coupling amount.

As one method for solving such inconvenience, correlation fluctuation processing units 7 ₁ ,... 7 _M are provided as shown in FIG. 4, and the received signal is amplitude-modulated with a random number for each channel, for example, to obtain the original received signal. In addition, Japanese Patent Application No. 7-50002, documents S. Shimauchi and S. Makino, generate a signal whose cross-correlation is constantly changing and reproduce it from each speaker and simultaneously input it to a multi-channel echo canceller. , “Stereo Projection Echo Canceller with True Echo Path Estimation,” Proc.ICASSP95, vol.5, pp. 3059-3062 (1995). Thereafter, according to Japanese Patent Laid-Open No. 2002-223829, a new adaptive algorithm for obtaining a correction vector of an adaptive filter from a signal in which the ratio of received signal information to additional signal information is reduced.・ An echo cancellation method has been proposed.

Even if the above-mentioned proposals improve the estimation performance of the multi-channel echo path, the following disadvantages occur.
When the difference between the estimated value of the echo transfer characteristic and the true value is viewed as a relative error of the filter coefficient, the behavior of the residual echo and the relative error is as shown in FIG. In a state where one speaker is speaking to the ground, the echo is almost eliminated and the error signal power becomes almost zero, but the relative coefficient error is saturated without converging. Therefore, the echo is not suddenly erased at the moment when the speaker changes on the ground.
If it is determined that the echo path has been sufficiently estimated with the echo almost erased and the attenuation by the attenuator is cancelled, the echo will not be erased suddenly at the moment when the speaker on the ground changes, and the worst If you lead to howling.

Therefore, in the present invention, information on the echo path estimation error of the multi-channel adaptive filter is extracted, and the attenuation applied to the receiving side or the transmitting side is controlled according to this information.
In the present invention, as shown in FIG. 6C, L _p taps in the adaptive filter constituting the pseudo-echo path are assigned to non-causal components (components before a certain time), and from the input signal up to time k to time k−L to predict the _p of the echo signal _{y ^ (k-L p)} . The filter coefficient vector of the pseudo echo path is omitted when time k is omitted.

  When the estimation of the echo path by the adaptive filter is completed and each pseudo echo signal has converged to a true echo signal, all the non-causal components of the pseudo echo path converge to 0, and the causal components of the pseudo echo path are true echoes. Converge to the path. In the middle of estimation, the difference between the causal component of the pseudo echo path and the true echo path is linked to the size of the non-causal component of the pseudo echo path. It is possible to know the estimated state of the route. Specifically, the acoustic echo canceling method according to the present invention is executed by the following acoustic echo canceling apparatus.

According to the present invention, in an acoustic echo canceling apparatus that cancels a multi-channel acoustic echo by a combination of an adaptive loss control unit and a multi-channel acoustic echo canceling unit, the acoustic echo canceling unit includes a pseudo-acoustic echo signal including a non-causal component. A pseudo acoustic echo signal generation unit that generates a non-causal component, an echo path estimation unit that includes a non-causal component, and a subtractor that calculates a difference between the pseudo acoustic echo signal including the non-causal component generated by the pseudo acoustic echo signal generation unit and the echo signal And a non-causal component and a causal component of the pseudo acoustic echo path estimated by the echo path estimation unit, and a consistency estimation unit that extracts an error of the echo path and the pseudo acoustic echo path as echo path estimation error information, An acoustic echo canceller characterized by controlling the attenuation amount of the adaptive loss control unit based on the echo path estimation error information extracted by the coincidence estimation unit Proposed.
According to the present invention, in the acoustic echo canceller, the acoustic echo canceler includes a canceling pseudo echo signal generator that does not include a non-causal component, and the canceling pseudo echo signal generator cancels only the causal component. The present invention proposes an acoustic echo canceller characterized by being applied.

  According to the present invention, a non-causal component is included in a multi-channel adaptive filter, the estimation state of the multi-channel adaptive filter is estimated from the size of the non-causal component, and the attenuation ratio on the receiving side or transmitting side is controlled according to the estimation. By doing so, howling of the multi-channel loudspeaker communication system is surely prevented regardless of the estimation state of the echo path by the multi-channel adaptive filter.

  The best mode for carrying out the present invention will be described below as a first embodiment.

A first embodiment of the present invention is shown in FIG. Here, description will be made using a multi-channel loudspeaker communication system including an M-channel regenerator and a one-channel sound collector.
The M channel reception signal is reproduced from the speakers 1 ₁ to 1 _M via the reception side attenuator 31, and the echo is collected by the microphone 2.
The M channel signal is simultaneously input to the pseudo echo signal generating unit 51 of the acoustic echo canceling unit 6 to generate a pseudo echo signal. The attenuator 52 obtains an error signal by subtracting the pseudo echo signal from the echo signal to cancel the echo. In the present invention, a delay unit 50 is provided in the acoustic echo canceller 6, and a non-causal component is generated by the delay of the delay unit 50.
The echo path estimation unit 53 updates the estimated value of the pseudo echo path from the received signal and the error signal.

In the present invention, the pseudo echo signal generation unit 51 and the echo path estimation unit 53 include a non-causal component, and the coefficient vector of the m-th channel pseudo echo path omits the time k.

により求める。ここではαはエコー信号の非因果成分の大きさを因果成分の大きさに換算するための定数であり、例えばα＝（Ｌ−Ｌ_ｐ）／Ｌ_ｐを用いる。δは０割りを防止するための正則化定数である。 In the coincidence estimation unit 54, the coincidence Q (k) between the echo signal and the pseudo echo signal is calculated from the magnitude of the non-causal component and the causal component, for example,

Ask for. Here, α is a constant for converting the magnitude of the non-causal component of the echo signal into the magnitude of the causal component. For example, α = (L−L _p ) / L _p is used. δ is a regularization constant for preventing 0 division.

The coincidence index Q (k) takes a value of 0 or more. In the state where the estimation by the adaptive filter is started and the echo is hardly erased and the coincidence between the echo path and the pseudo echo path is low, Q (k) takes a value near 1. When the estimation by the adaptive filter is completed and the pseudo echo path and the echo path are completely coincident, all the non-causal components of the pseudo echo path converge to 0, and Q (k) = 0.
The loss control unit 33 determines the transmission state from the received signal and the error signal output from the subtractor 52 of the acoustic echo canceling unit 6. The attenuation ratio R (k) applied to transmission or reception is controlled in accordance with the consistency index Q (k) obtained by the consistency estimation unit 54.

を用いることができる。減衰比Ｒ（ｋ）は０から１の値をとる。ただし、ＬはｄＢで設定された初期減衰量であり、エコーキャンセラの起動直後に適応フィルタによりエコーがほとんど消去されない状態で、送話もしくは受話に適用される減衰量である。通常Ｌ＝−２０〜−３０ｄＢに設定することで起動直後のハウリングを防止できる。
受話状態と判定されたときには、送話側減衰器３２で誤差信号に減衰比Ｒ（ｋ）を適用する。また送話状態と判定されたときには、受話側減衰器３１で受話信号に減衰比Ｒ（ｋ）を適用する。 As a control method of the damping ratio R (k), for example,

Can be used. The damping ratio R (k) takes a value from 0 to 1. However, L is an initial attenuation set in dB, and is an attenuation applied to transmission or reception in a state where the echo is hardly erased by the adaptive filter immediately after activation of the echo canceller. Usually, howling immediately after startup can be prevented by setting L = -20 to -30 dB.
When it is determined that the voice is received, the transmission side attenuator 32 applies the attenuation ratio R (k) to the error signal. When the transmission state is determined, the reception side attenuator 31 applies the attenuation ratio R (k) to the reception signal.

In the case of a multi-channel loudspeaker communication system with M channel reproduction and N channel sound pickup, N M channel echo cancelers 6 _{1 to} 6 _N are arranged in parallel as shown in FIG. Attenuation ratios R ₁ (k)... R _N (k) are obtained at the erasing unit. The attenuation ratio R (k) applied at the reception side attenuator or the transmission side attenuator is, for example, from the attenuation ratios R ₁ (k)... R _N (k).
R (k) = MIN [R ₁ (k) ... R _N (k)]
It can be decided by.

とし、フィルタ長Ｌ、非因果成分Ｌ_Ｐ（＞１）タップとする。
チャネル間相関変動処理部７では、受話信号Ｕ_ｍ（ｋ）（ｍ＝１…Ｍ）に例えばランダムに変化する付加信号Ｕ_ｍ ^ＡＤＤ（ｋ）を付加して、スピーカ再生信号Ｘ_ｍ（ｋ）を生成する。
ｘ_ｍ（ｋ）＝ｕ_ｍ（ｋ）＋Ｕ_ｍ ^ＡＤＤ（ｋ）
（ｍ＝１…Ｍ） A second embodiment of the present invention will be described with reference to FIG. Similar to the first embodiment, the pseudo echo signal generation unit 51 and the echo path estimation unit 53 include a non-causal component that is enclosed in a block at the same time as the causal component.

And a filter length L and a non-causal component L _P (> 1) tap.
The inter-channel correlation fluctuation processing unit 7 adds, for example, an additional signal U _m ^ADD (k) that randomly changes to the received signal U _m (k) (m = 1... M), and reproduces the speaker reproduction signal X _m (k). Is generated.
x _m (k) = u _m (k) + U _m ^ADD (k)
(M = 1 ... M)

である。適応フィルタの更新には、上記のような時間領域の適応アルゴリズム以外に、周波数領域の適応アルゴリズムをもちいることもできる（特開２００２−２２３１８２号公報参照）。
実施例１と同様に、一致性推定部５４においてエコー信号と擬似エコー信号の一致性が推定され、この推定値にもとづいて損失制御部３３で減衰比を求めて適応損失制御を行う。 The pseudo echo is generated in the same manner as in the first embodiment. Output from the inter-channel correlation fluctuation processing unit 7. The correction vector generation signal Z _m (k) is obtained from the attenuated reception signal and the additional signal.
z _m (k) = βu _m (k) + U _m ^ADD (k)
(M = 1 ... M, 0 <β <1)
Is generated as follows. Adaptive filter, an error signal between the echo signal L _P sample delay
e (k−L _p ) = y (k−L _p ) −y ^ (k−L _p )
(M = 1 ... M)
For example,

It is. For updating the adaptive filter, in addition to the time domain adaptive algorithm as described above, a frequency domain adaptive algorithm can also be used (see JP 2002-223182 A).
As in the first embodiment, the coincidence estimation unit 54 estimates the coincidence between the echo signal and the pseudo echo signal, and the loss control unit 33 obtains the attenuation ratio based on the estimated value and performs adaptive loss control.

A third embodiment of the present invention will be described with reference to FIG. The third embodiment corresponds to the acoustic echo canceling method proposed in claim 3 of the present invention.
The M channel reception signal that has passed through the reception side attenuator is reproduced from the speakers 1 ₁ to 1 _M , and the echo is collected by the microphone 2. The M channel signal is simultaneously input to the pseudo echo signal generator 51 of the echo canceler 6 to generate a pseudo echo signal. The error signal obtained by subtracting the pseudo echo signal from the echo signal in the subtractor 52 is used for updating the pseudo echo signal in the echo path estimation unit 53. The M channel signal is also input to the erasing pseudo echo signal generation unit 55 to generate a pseudo echo signal. The subtractor 56 subtracts the pseudo echo signal from the echo signal to obtain an error signal, thereby canceling the echo.

The pseudo echo signal generation unit 51 and the echo path estimation unit 53 are configured to include a non-causal component, and the erasing pseudo echo signal generation unit 55 is configured not to include a non-causal component. Only the causal component is transferred from the echo path estimation unit 53 to the erasing pseudo echo signal generation unit 55. As a result, it is possible to transmit the transmission signal to the ground without being affected by the delay inserted to obtain the non-causal component.
As in the first embodiment, the coincidence estimation unit 54 estimates the coincidence between the echo signal and the pseudo echo signal from the magnitude of the non-causal component obtained by the echo path estimation unit 53, and uses the estimation result as the loss control unit. Pass to 33.

Here, the description has been made based on FIG. 7 of the first embodiment, but this method can also be applied to FIG. 8 of the first embodiment and FIG. 9 of the second embodiment.
The acoustic echo canceling apparatus according to the present invention described above is realized by causing a computer to execute the acoustic echo canceling program according to the present invention.
The acoustic echo cancellation program according to the present invention is written in a computer-readable program language, recorded on a recording medium such as a magnetic disk or a CD-ROM, and installed in the computer from these recording media or through a communication line. Installed.
The acoustic echo cancellation program installed in the computer is decoded by a CPU provided in the computer, and the above-described acoustic echo cancellation is executed.

  The acoustic echo cancellation method and apparatus according to the present invention provide echo cancellation especially for multi-channel signals. As a result, the present invention can be utilized in a conference system with multi-channel playback and high presence.

The connection diagram for demonstrating the fundamental structure of a voice call system. The block diagram for demonstrating the prior art. The block diagram for demonstrating the acoustic echo cancellation apparatus multi-channeled by the prior art. The block diagram for demonstrating the example which performed the correlation fluctuation process between channels to the prior art. The figure for demonstrating the inconvenience at the time of multi-channeling by the prior art. The figure for demonstrating the impulse response of the echo path | route used as the principal part of this invention, and the causal component and non-causal component of an adaptive filter. The block diagram for demonstrating 1st Example of this invention. FIG. 8 is a block diagram showing an example in which the embodiment shown in FIG. 7 is applied to a communication conference system including an M channel reproduction system and an N channel sound collection system. The block diagram for demonstrating 2nd Example of this invention. The block diagram for demonstrating 3rd Example of this invention.

Explanation of symbols

1 Speaker 33 Loss Control Unit 1 _I to 1 _M Speaker 41 Pseudo Echo Signal Generation Unit
2 Microphone 42 Subtractor 3, 4 Transmission path 43 Echo path estimation unit
5 Adaptive loss control unit 51 Pseudo echo signal generation unit including non-causal components
6 Acoustic echo canceller 52 Subtractor
7 Inter-channel correlation fluctuation processing unit 53 Echo path estimation unit including non-causal component 31 Receiving side attenuator 54 Consistency estimating unit 32 Transmitting side attenuator

Claims (6)

  In an acoustic echo cancellation method that cancels multi-channel acoustic echoes by combining adaptive loss control means and multi-channel acoustic echo cancellation means, the multi-channel adaptive filter includes non-causal components and extracts echo path estimation error information from this An acoustic echo canceling method characterized in that the attenuation amount on the receiving side or transmitting side in the adaptive loss control means is controlled by the extracted echo path estimation error information.   2. The acoustic echo canceling method according to claim 1, wherein the echo path estimation error information uses a level of path matching between the echo path and the pseudo echo path.   3. The acoustic echo canceling method according to claim 1, wherein the multi-channel acoustic echo canceling unit is provided with a canceling pseudo echo signal generating unit not including a non-causal component, and the canceling pseudo echo signal generating unit. An acoustic echo canceling method characterized by performing acoustic echo canceling of only causal components by the method described above. In the acoustic echo canceller that cancels the multi-channel acoustic echo by combining the adaptive loss control unit and the multi-channel acoustic echo canceler,
The acoustic echo canceling unit generates a pseudo acoustic echo signal generation unit that generates a pseudo acoustic echo signal including a non-causal component, an echo path estimation unit including a non-causal component, and the non-causal component generated by the pseudo acoustic echo signal generation unit An error between the echo signal and the pseudo acoustic echo signal is calculated from a subtractor for obtaining a difference between the pseudo acoustic echo signal including the echo signal and the non-causal component and the causal component of the pseudo acoustic echo signal estimated by the echo path estimation unit. Consistency estimation unit to extract as echo path estimation error information,
An acoustic echo canceling apparatus that controls the attenuation amount of the adaptive loss control unit based on the echo path estimation error information extracted by the coincidence estimation unit.   5. The acoustic echo canceling apparatus according to claim 4, wherein an erasing pseudo-echo signal generation unit that does not include a non-causal component is provided in parallel with the acoustic echo erasing unit, and an echo signal having only a causal component is generated by the erasing pseudo-echo signal generation unit. An acoustic echo canceling device characterized by performing canceling. 5. An acoustic echo canceling program that is written in a computer-readable program language and causes the computer to function as the acoustic echo canceling device according to claim 3 or 4.

Images