Method for adaptive control of multichannel acoustic echo cancellation system and device therefor

Abstract

The invention concerns a method for adaptive control of multichannel acoustic echo cancellation system comprising N speakers HPi (i=1, 2, N), N being an integer not less than 2, and M microphones MCj (j=1, 2, M), M being an integer not less than 1. The system comprises N×M identification filters Fij with variable coefficients, the identification filter Fij enabling to estimate the acoustic coupling between the speaker HPj and the microphone MCj based on a pitch μ adapting the variable coefficients. The pitch μ adapting the variable coefficients is based on the presence or absence of signal in the speakers HPk (k≠i)?. The invention is applicable to the hands-free use of communication tools.

Description

TECHNICAL DOMAIN AND PRIOR ART

[0001] This invention relates to a method for adaptive control of a multi-channel echo cancellation system and a device for implementing the method.

[0002] The invention is applicable to hands-off operation of communication tools, for example such as mobile telephones, personal computers (PCs) and more generally any type of device present in audio and/or video workstations in which audio communication is made using several loudspeakers at a distance from the participants.

[0003] Acoustic echo is a major obstacle to smooth hands-off operation of communication tools. Acoustic echo is the result of a signal which is emitted by a loudspeaker, and is picked up by a microphone either directly or by reflection. Many studies have been carried out on the problem of acoustic echo, both for single-dimensional and for multi-dimensional cases.

[0004] In the single-dimensional case, there is only one sound pick-up signal and only one sound reproduction signal, even if the sound pick-up signal is picked up by several microphones and the sound reproduction signal is reproduced on several loudspeakers.

[0005] In the multi-dimensional case, an echo cancellation system comprises N signal reception channels, each comprising a loudspeaker Hpi (i=1, 2, . . . , N) and M sound pick-up channels, each sound pick-up channel comprising a microphone MCj (j=1, 2, . . . , M). This type of echo cancellation system is shown in FIG. 1. It comprises N×M acoustic channels and consequently N×M echo cancellation devices H (i, j). Each echo cancellation device H(i, j) comprises an identification filter that estimates acoustic coupling between the loudspeaker HPi and the microphone MCj. Apart from the problem of real time operation of such a system, due to the required calculation power, another problem arises related to convergence of the identification filter control algorithm.

[0006] Many studies have been carried out on the problem of convergence of the identification filter control algorithm. For example, convergence methods include the method based on the addition of random noise or the method based on introduction of a non-linear function to decorrelate signals to be processed. This method is described in U.S. Pat. No. 5,828,756 by Benesty et al. issued in the United States on Oct. 27, 1998 entitled “Stereophonic acoustic echo cancellation using non-linear transformations”.

[0007] Convergence methods according to prior art have many disadvantages, including the long calculation times.

[0008] The invention does not have the disadvantages mentioned above.

PRESENTATION OF THE INVENTION

[0009] The invention relates to a method for adaptive control of a multi-channel echo cancellation system comprising N loudspeakers HPi (i=1, 2, . . . , N), where N is an integer greater than or equal to 2, and M microphones MCj (j=1, 2, . . . , M), where M is an integer greater than or equal to 1, the system comprising N×M identification filters Fij with variable coefficients, the identification filter Fij being used to check acoustic coupling between the loudspeaker HPi and the microphone MCj under the action of control information, the control information being calculated using an adaptive algorithm based on an error signal between a signal detected by the microphone MCj and a reference signal that includes the estimated signal output from the identification filter Fij, and a variable coefficients adaptive step. The reference signal also comprises a signal equal to the sum of P estimated supplementary signals ŷk,jk≠i (k=1, 2, . . . , i−1, i+1, . . . , P+1) output from P identification filters Fkjk≠i, where P is an integer between 1 and N-1, and the variable coefficients adaptive step depends on whether or not there is a signal present on the P loudspeakers HPk (k≠i).

[0010] The invention also relates to a device for adaptive control of a multi-channel echo cancellation system comprising N loudspeakers HPi (i=1, 2, . . . , N), where N is an integer greater than or equal to 2, and M microphones MCj, where M is an integer greater than or equal to 1, the system comprising an identification filter Fij with variable coefficients to estimate acoustic coupling between the loudspeaker HPi and the microphone MCj, the identification filter being controlled by control information output from an update unit controlled by an error signal between a signal detected by the microphone and a reference signal that includes the estimated signal output from the identification filter Fij, and by a variable coefficients adaptive step. The device comprises means of adding the estimated signals ŷk,j k≠i(k=1, 2, . . . , i−1, i+1, . . . , P+1) output from P identification filters Fkj k≠i, to the reference signal, where P is an integer between 1 and N-1 and means of modifying the value of the variable coefficients adaptive step depending on whether or not there is a signal present on the P loudspeakers HPk (k≠i).

[0011] As will become clear later, in one preferred embodiment of the invention the number P is equal to N-1.

BRIEF DESCRIPTION OF THE FIGURES

[0012] Other characteristics and advantages of the invention will become clear after reading the preferred embodiment with reference to the attached figures in which:

[0013] FIG. 1 shows a principle diagram of a multi-channel echo cancellation system;

[0014] FIG. 2 shows an adaptive control device for a multi-channel echo cancellation system according to a first embodiment of the invention;

[0015] FIG. 3 shows an adaptive control device for a multi-channel echo cancellation system according to a second embodiment of the invention;

DETAILED DESCRIPTION OF EMBODIMENTS OF THE INVENTION

[0016] FIG. 1 has already been described, therefore there is no point in describing it again.

[0017] FIG. 2 shows an adaptive control device of the multi-channel echo cancellation system according to a first embodiment of the invention.

[0018] The control device in FIG. 2 is the control device according to the invention that enables echo cancellation between the microphone MCj and the loudspeaker HPi.

[0019] The control device comprises an identification filter Fij with variable coefficients, an update unit 1, a means 2 of calculating the variable coefficients adaptive step μ, N voice activity detectors DAV(xi) (i=1, 2, . . . N), a voice activity detector DAV(yj), a subtractor 3 and an adder 4. Each voice activity detector indicates whether or not a voice signal is present, as a function of a decision threshold. Thus, the signal d(xi) output from the voice activity detector DAV(xj) indicates whether or not there is a signal present in the loudspeaker HPi. Similarly, the signal d(yj) output from the detector DAV(yj) indicates whether or not there is a signal present in the microphone MCj.

[0020] The identification filter Fij is a programmable filter with a finite pulse response for which the coefficients must be adapted. The update unit 1 adapts the filter coefficients based on the signal xi present in the loudspeaker HPi, the calculated step μ of the variable coefficients and an error signal e. For example, the update unit 1 uses the Normalized Least Mean Squares (NLMS) algorithm or the order 2 Affine Projection Algorithm (APA2).

[0021] The update unit 1 updates the filter Fij when the following conditions are satisfied:

[0022] the detector DAV(xi) detects a voice activity,

[0023] the detector DAV(yj) does not detect any voice activity,

[0024] the detectors DAV(Xk, k≠i) do not detect any voice activity.

[0025] According to the first embodiment of the invention, the value of the step μ is equal to zero if any one of the signals d(xk), k≠i, indicates that there is a signal present on a loudspeaker HPk. If no signal is detected, the adaptive step μ is chosen to be the optimum considered for the adaptive algorithm used by the update unit. For example, the coefficient μ is then equal to 0.33 for the NLMS algorithm or the APA2 algorithm.

[0026] The error signal e is equal to the difference between the signal yj present in the microphone MCj and a reference signal. The reference signal is composed of the estimated signal ŷi,j output from the identification filter Fij and all estimated signals ŷk,j k≠i output from the N-1 identification filters Fk,j. Consequently, the signals ŷk,j k≠i output from the N-1 identification filters Fk,j are summated by the adder 4.

[0027] The error signal is written as follows: 1$e=y_j-\left(\hat{y}i,j+\sum _k{\hat{y}}_{k,jk\ne i}\right)$

[0028] The device according to the invention advantageously reduces the calculation time for the update step since fixing the step μ to the value zero in the presence of a signal on any of the loudspeakers stops the convergence calculation.

[0029] FIG. 3 shows an adaptive control device for the multi-channel echo cancellation system according to a second embodiment of the invention.

[0030] According to the second embodiment of the invention, the coefficients adaptive step μ is calculated based on the principle described in the French patent application entitled “Procédé et dispositif d'identificatipn adaptative et annuleur d'écho adaptatif incluant un tel dispositif” (“Adaptive identification method and device and adaptive echo canceller comprising such a device”), deposited in France on Sep. 13, 1995 and published as No. 2 738 695. The control device comprises an identification filter Fij with variable coefficients, an update unit 5, a means 6 of calculating the variable coefficients adaptive step μ, a voice activity detector DAV. (xi) on the loudspeaker HPi, an estimator 7 of the energy Xi of the signal xi, an estimator 8 of the energy Yj of the signal yj, an estimator 9 of the weighted energy 2$\sum _k$

[0031] dkXk,k≠i of signals xk (k≠i) output from the N-1 loudspeakers HPk (k=1, 2, . . . , i−1, i+1, . . . , N), a subtractor 3 and an adder 4.

[0032] The update unit 5 updates the filter Fij when the detector DAV (xi) detects voice activity.

[0033] The variable coefficients adaptive step μ is then calculated by the following expression: 3$\mu =\frac{a_{i}Xi}{b_i{X}_i+C_j{Y}_j+\sum _k\left(d_k{X}_{k,k\ne i}\right)}$

[0034] where ai, bi, cj and dk (k≠i) are positive coefficients. As a non-limitative example, the coefficient ai may be equal to 1, the coefficient bi may be equal to 3, the coefficient cj may then be between 10 and 100 (depending on the acoustic environment conditions) and the coefficients dk may be approximately of the same order of magnitude as the coefficient cj.

[0035] Advantageously, the coefficients dk can adjust the control of the step as a function of the number of loudspeaker channels. If there is a voice activity on at least one of the loudspeaker channels, the coefficient μ tends towards zero and adaptation is stopped.

[0036] One particular advantage of this second embodiment of the invention is that the coefficient μ can be made to tend continuously towards zero as the energy in the signals increases. This advantage does not exist when voice activity detectors are used since adaptation can take place with parasite signals as long as the decision threshold has not been reached. According to the second embodiment of the invention, it is particularly easy to prevent mismatching of the filter.

[0037] Regardless of its embodiment, the invention is advantageously applicable to any type of adaptive algorithm.

Claims

1. Method for adaptive control of a multi-channel echo cancellation system comprising N loudspeakers HPi (i=1, 2, . . . , N), where N is an integer greater than or equal to 2, and M microphones MCj (j=1, 2, . . . , M), where M is an integer greater than or equal to 1, the system comprising N×M identification filters Fij with variable coefficients, the identification filter Fij being used to estimate acoustic coupling between the loudspeaker HPi and the microphone MCj under the action of control information, the control information being calculated using an adaptive algorithm based on an error signal (e) between a signal detected by the microphone MCj (yj) and a reference signal that includes the estimated signal (ŷij) output from the identification filter Fij, and a variable coefficients adaptive step (μ), characterized in that the reference signal also comprises a signal equal to the sum of P estimated supplementary signals ŷk,jk≠i (k=1, 2, . . . , i−1, i+1, . . . , P+1) output from P identification filters Fkj(k≠i), where P is an integer between 1 and N-1, and the variable coefficients adaptive step (μ) depends on whether or not there is a signal present on the P loudspeakers HPk.

2. Method according to claim 1, characterized in that the variable coefficients adaptive step (μ) is equal to zero if there is a signal present on at least one of the P loudspeakers HPk.

3. Method according to claim 2, characterized in that, when the adaptive algorithm is the Normalized Least Mean Squares (NLMS) algorithm or the order 2 Affine Projection Algorithm (APA2), the variable coefficients adaptive step (μ) is equal to 0.33 if no signal is detected on the M loudspeakers HPk.

4. Method according to claim 1, characterized in that it comprises: a step to estimate the energy Xi of the acoustic signal output by the loudspeaker HPi, a step to estimate the energy Yj of the signal received by the microphone MCj, a step to estimate the weighted energy 4∑k dkXk,k≠of signals output from P loudspeakers HPk, where P is an integer between 1 and N-1, and in that the variable coefficients step μ is calculated by the following expression: 5μ=ai⁢X⁢ ⁢ibi⁢Xi+Cj⁢Yj+∑k⁢(dk⁢Xk,k≠i)where ai, bi, cj and dk (k≠i) are positive coefficients.

5. Method according to claim 4, characterized in that: ai=1, bi=3, c3 is between 10 and 100, and dk is between 10 and 100.

6. Device for adaptive control of a multi-channel echo cancellation system comprising N loudspeakers HPi (i=1, 2, . . . , N), where N is an integer greater than or equal to 2, and M microphones MCj (j=1, 2, . . . , M), where M is an integer greater than or equal to 1, the system comprising a variable coefficients identification filter Fij being used to estimate acoustic coupling between the loudspeaker HPi and the microphone MCj, the identification filter being controlled by control information output from an update device (1, 5) controlled by an error signal (e) between a signal detected by the microphone (yj) and a reference signal that includes the estimated signal (ŷij) output from the identification filter Fij, and by a variable coefficients adaptive step (μ), characterized in that it comprises means of adding the estimated signals (ŷk,j k≠i(k=1, 2, . . . , i−1, i+1, . . . , P+1)) output from P identification filters Fkj (k·i), to the reference signal, where P is an integer between 1 and N-1, and means of modifying the value of the variable coefficients adaptive step (μ) depending on whether or not there is a signal on P loudspeakers HPk.

7. Device according to claim 6, characterized in that the means of modifying the value of the step (μ) comprise P voice activity detectors DAV(xk), the detector DAV(xk) indicating whether or not there is a signal on the loudspeaker HPk.

8. Device according to claim 6, characterized in that the means of modifying the value of the step (μ) comprise first means (7) of estimating the energy Xi of the acoustic signal output by the loudspeaker HPi, second means (8) of estimating the energy yj of the signal received by the microphone MCj, third means (9) of estimating the weighted energy

9. Device according to claim 8, characterized in that: ai=1 bi=3 cj is between 10 and 100, and dk is between 10 and 100.