WO2007003579A1

WO2007003579A1 - Device for cancelling electric or acoustic echo, corresponding method and computer programme

Info

Publication number: WO2007003579A1
Application number: PCT/EP2006/063697
Authority: WO
Inventors: Grégoire Le Tourneur; David Deleam
Original assignee: France Telecom
Priority date: 2005-07-01
Filing date: 2006-06-29
Publication date: 2007-01-11

Abstract

The invention concerns a device for cancelling an echo present in a return channel (14), and corresponding to a direct channel (11) of an audio communication system, comprising echo processing means including means (16) for adaptive filtering of the direct channel, delivering an estimation of the echo to be subtracted from the return channel (14). The invention is characterized in that said device comprises means for controlling (26) the processing means taking into account a comparison between a first information representing the energy in output of the processing means and a second information representing the energy of the return channel (14).

Description

Device for canceling an electric or acoustic echo, corresponding method and computer program. 1. Field of the invention

The field of the invention is that of digital communications, and in particular that of the processing of audio signals in a communication system.

More specifically, the invention relates to an echo cancellation technique for attenuating, or at least reducing, in a return path, electrical or acoustic echo components corresponding to a direct path, even in case of change of the echo path.

The invention finds particular applications in audio communication systems such as telephone terminals, VoIP gateways ("Voice over IP", in French "Voice over Internet Protocol"), call centers, etc. 2. Solutions of the prior art and their disadvantages

Already known in the field of telecommunications, a technique for canceling at least partly an echo, implementing an adaptive filtering.

This technique makes it possible in particular to solve the problems of electrical echoes, resulting for example from line transformers (especially two-wire / four-wire), or the acoustic echo problems, resulting for example from the hands-free functionality proposed in the different types of terminals.

An example of this echo cancellation technique, in a telephony or teleconferencing application between one or more remote party (s) and a local correspondent, is illustrated in FIG.

A direct or reference signal 11, carried by a direct channel 11, is broadcast on a loudspeaker 12, and a return signal 14, carried by a return channel 14, is returned by a microphone 13. Such a signal of return 14 optionally includes echo components of the reference signal 11 and a local speech signal. The echo path 15 then corresponds to the materialization of all the reflections undergone by the direct signal 11 before reaching the microphone 13.

We note that the channels have, for the sake of simplification, the same names and reference signs as the signals they carry.

The cancellation of the echo by adaptive filtering according to this technique of the prior art is based on the identification of the echo path 15, and requires the modeling of this echo path 15 using an adaptive filter 16.

An adaptive filtering algorithm, for example of the NLMS (Normalized Least Mean Square) type, MCR (Least Squares Fastest Square), or AP A2 (Affine Projection Algorithm). order 2), in particular makes it possible to adjust the coefficients of the adaptive filter according to a recursive equation of the type:

H (n + 1) = H (n) + μ (n) * error (n) * f (x (n)) where: - H (n) represents the adaptive filter;

x (n) represents the reference signal; f (x (n)) is a function of the vector of the last L samples of the excitation signal (L integer), which function varies according to the families of algorithms considered;

μ (n) represents the adaptation step;

- error (n) represents the distance between the real echo and the estimated echo, ie the distance between the feedback signal y (n), composed of the echo (n) echo signal and the speech signal (n), and the estimated estimated echo signal (n) at the output of the adaptive filter. In other words, the error signal (n) is defined by error (n) = y (n) - estimated (n), with only local speech y (n) = speech (n), in case of single echo y (n) = echo (n), and in case of double speech y (n) = echo (n) + speech (n).

However, a disadvantage of this prior art technique is that such an adaptive filter-based system is highly recursive, and therefore requires appropriate control means to ensure its stability.

Consequently, the adaptation of the coefficients according to the preceding equation must be done only in the presence of echo alone (y (n) = echo (n)) to estimate correctly the actual echo path, ie in the absence of simultaneous local speech.

In addition, in real-life echo canceller situations, there are situations where the echo disappears completely from the input of the system: for example, when a telephone terminal changes from hands-free mode In "handset" mode, or when certain telephone lines of a telephone set connected to several lines in parallel are released (professional terminals, call center, etc.).

If the echo path change information is not communicated to the echo canceller, the echo canceller remains stuck in the state corresponding to the pre-change state, i.e. of the filter remain adapted to a certain echo path while this path has changed. This case can occur because of the architecture of the application.

According to the prior art, techniques for detecting an echo path change based on the use of voice activity detectors (VAD) have been proposed.

These conventional techniques propose to use at least one voice activity detector on the feedback signal y (n) to determine the state of the system (echo only, local speech only, double talk or silence), and to apply a control strategy. adaptation according to the determined state.

Thus, when the phenomenon of disappearance of the echo occurs, the absence of signal on the return channel causes a "non vocal activity" on the detectors, and adaptation is not done. Indeed, in the absence of signal y (n) on the microphone, the voice activity detectors used for the control of the algorithm freeze the adaptation of the filter, and the system remains in the state: the coefficients of the adaptive filter are no longer suitable.

More precisely, if the system has correctly converged before the physical removal of the echo source, the adaptive filter represents a filter very close to the transfer function of the echo source. The estimated signal (n), obtained by convolution of the reference signal x (n) with the coefficients of the adaptive filter, is therefore close to the signal y {n) = echo (n) in the absence of speech, and the error signal (n) is low to very low: estimated (n) = 2_, x (n - k) • h (k) for k varying from 0 to N, where N represents the length of the adaptive filter; estimated (n) ≈ echo (n); error {n) = echdyή) - estimated (n) ≈ 0. If the echo source disappears, the equations become: echc {ή) ≈ 0; error (n) ≈ - estimated (n); error (n) = -2_, x (n - k) - h (k).

Under these conditions, the system no longer behaves as an echo canceller, but as an echo generator in the presence of the direct signal x (n). The user then perceives an echo on the return path, while the echo source has disappeared.

Moreover, the use of voice activity detectors for determining the state of the system is not optimal, since these detectors confuse in certain cases the state of double speech and the state of echo alone, and diverge the adaptive filter, turning it into a random noise generator.

It may also be noted that in certain variants of the techniques of the prior art, the blocking phenomenon disappears of itself after a certain time if the filter converges to a new transfer function ("null" function since no more echo). However, this convergence requires a certain delay, during which the echo cancellation device behaves like a noise generator.

3. Objectives of the invention

The invention particularly aims to overcome these disadvantages of the prior art. More precisely, an objective of the invention is to propose an echo cancellation technique making it possible to improve the intelligibility and interactivity of a conversation by rapidly detecting a change in the echo path, and in particular by quickly detecting the suppression of an echo. Thus, an objective of the invention is to propose such a technique that makes it possible to eliminate, or at least reduce, the audible part of a residual echo, without causing perceptible degradation on a local speech signal, and without generate random noise when changing the echo path.

In particular, the invention aims to provide such a technique for stabilizing an echo cancellation system in case of change of the echo path.

The object of the invention is also to propose such a technique that is simple to implement and does not require the addition of voice activity status indicator (s), while presenting improved performances with respect to state of the art techniques.

4. Presentation of the invention

These objectives, as well as others which will appear later, are achieved by means of an echo cancellation device present in a return channel, and corresponding to a direct channel of a communication system. audio, comprising echo processing means comprising adaptive filtering means of the direct channel, said echo processing means providing an estimate of the echo to be subtracted from the return channel.

According to the invention, such a device comprises means for controlling the processing means taking into account a comparison between a first information representative of the energy output of the processing means and a second information representative of the energy of the channel. back.

Thus, the invention is based on a completely new and inventive approach to the control of the echo processing means, making it possible to quickly adapt the parameters of such an echo cancellation device in the event of detection of a change of the echo path. More specifically, the processing means comprise at least one adaptive filter, and may for example comprise a switch controlling such an adaptive filter.

According to the invention, by comparing the energy output of the processing means and the energy carried by the return path, it can be deduced that the echo source has been modified, and that consequently the processing means of echo are no longer suitable for the new echo path.

The control means then act on the processing means, in order to adapt the device to the new echo path. By quickly detecting a change in the echo source, the technique according to the invention thus makes it possible to prevent the echo cancellation device from behaving like a noise generator.

Advantageously, the control means determine a ratio between the first and second information, and the comparison compares this ratio to a first threshold.

Thus, by comparing the ratio between the energy output of the processing means and the energy carried by the return path to a first threshold, and using a criterion of exceeding this threshold, the invention makes it possible to quickly detect a change in the echo path, including the suppression of an echo, and possibly to act on the means of echo processing.

Preferably, the control means comprise delay means, triggered when the ratio is greater than the first threshold.

These delay means make it possible in particular to delay the activation of an operation on the processing means, avoiding false alarms, which may occur when the energy output of the processing means exceeds the energy carried by the return path. for a very short time.

According to the invention, it is therefore verified that the ratio between the energy output of the processing means and the energy carried by the return channel is greater than the first threshold for at least a certain period defined by the delay means. These delay means may comprise at least one counter. Advantageously, the control means deliver an echo path change information when the delay means exceeds a second threshold. Thus, the decision of detecting a change in the echo path is taken when the ratio between the energy at the output of the processing means and the energy carried by the return path is kept higher than the first threshold up to that the delay means exceeds a second threshold.

The control means then act on the echo processing means. Preferably, the control means act on at least one coefficient of the filtering means, by means of the processing means, in order to adapt them to the new echo path. The processing means may in particular control at least one switch for adjusting the coefficients of the adaptive filter. The control means also make it possible to inhibit the filtering means by calling a function of the echo cancellation device.

The first threshold may in particular be predetermined or adapted according to a first control information representative of a state of the device. More precisely, this first threshold may be fixed ("absolute fixed"), that is, it is fixed and does not change over time. This threshold can also be predetermined according to a first control information representative of the state of the device, taking for example account of the nominal value of the coupling

(ie the level of the echo return), and / or the current value of the delay means, and / or the number of false alarms. Finally, this first threshold can also be adapted over time.

The second threshold may also be predetermined or adapted according to a second control information representative of a state of the device. More precisely, this second threshold can also be fixed or adaptive, absolute or dependent on a second control information representative of the state of the device, taking into account the current value of the delay means, and / or the number of false alarms.

The invention also relates to a method for canceling an echo present in a return channel, and corresponding to a direct channel of an audio communication system, comprising an echo processing step implementing a substep adaptive filtering of the direct channel, said echo processing step providing an estimate of the echo to be subtracted from the return channel.

According to the invention, such a method comprises a step of controlling the processing step, taking into account a comparison between a first information representative of the energy at the output of the processing step and a second information representative of the energy of the return path.

The invention also relates to a computer program product downloadable from a communication network and / or stored on a computer readable medium and / or executable by a microprocessor, comprising program code instructions for the implementation of the method of cancellation of an echo.

5. List of figures

Other features and advantages of the invention will emerge more clearly on reading the following description of a preferred embodiment, given as a simple illustrative and nonlimiting example, and the appended drawings, among which: FIG. 1 , already commented upon in connection with the prior art, presents a conventional echo cancellation device comprising an adaptive filter, used for example in a telephony or teleconferencing application between one or more distant correspondent (s) and a local correspondent ; FIG. 2 illustrates an echo cancellation device comprising control means for echo processing means according to the invention. 6. Description of an embodiment of the invention The general principle of the invention is based on a new and original approach to the detection of a change in an echo source, in an echo cancellation system implementing an adaptive filtering technique. Such a technique notably makes it possible to dispense with the system state indicator (s), for example of the voice activity detector (s), which makes it possible to increase the reliability of the system and to adapt rapidly echo processing means to a new echo path.

More specifically, such a technique is based on the comparison of a first information representative of the energy output of the echo processing means, which notably comprise an adaptive filter, to a second piece of information representative of the energy on a signal path. return, which carries a local speech signal alone, echo only, or double speech: if the ratio between the first and second information is greater than a first threshold, a change of the echo path is detected. The proposed solution makes it possible in particular to detect the disappearance of a physical echo at the input of the system, and when such a change has been detected, to perform at least one operation aimed at restoring a stable and determined state of the system, by example by returning the adaptive filter in a stable and determined state. With reference to FIG. 2, an echo cancellation device comprising means for controlling the echo processing means according to the invention is presented.

As indicated previously with respect to the prior art, a return channel 14 carrying a feedback signal y (n) 14, possibly comprising echo echo components (n) corresponding to a direct channel 11 and a local speech signal. speech (n), enter the device according to the invention.

In this embodiment, it is considered that the echo processing means comprise only adaptive filtering means, and that the control means comprise first (respectively second) means for determining a first (respectively second) information. representative of the energy of a signal at the output of the echo processing means (respectively of the energy carried by the return channel), and means for comparing the first and second information. It is understood that the processing means could also comprise at least one switch, or any other means for adjusting the coefficients of the filtering means.

In order to model the echo path corresponding to the direct channel 11, the reference signal x (n) 11 is filtered by means of an adaptive filter H (n) 16, delivering an estimated estimated echo signal (n) 24. The estimated estimated echo signal (n) 24 is then subtracted from the feedback signal y (n) 14, delivering an error signal error (n) representing the distance between the real echo and the estimated echo .

We obtain the following equations: estimated (n) = 2_, x (n-k) - h (k) for k varying from 0 to N; y (n) = echo (n) + speech (n); error '(n) = y {n) - estimated {n). The technique according to the invention is based in particular on the following observations:

1. When the system has not yet converged, that is to say that the coefficients of the filter 16 are not adapted to cancel the echo, the estimated signal (n) 24 is close to zero, and the signal error (n) is close to the signal y (n) 14, in all situations of the return signal (echo only, local speech alone or double speech).

2. When the system has converged, that is to say that the transfer function of the adaptive filter is substantially equal to the transfer function of the echo source: - in echo only (y (n) = echo (n)), the estimated signal (n) 24 is close to the echo signal (n), and the error signal (n) is close to zero;

in the situation of only local speech (y (n) = speech (n)), the estimated signal (n) 24 is close to zero, and the error signal (n) is close to the speech signal (n); in a double speech situation (y (n) = echo (n) + speech (n)), the energy of the estimated signal (n) 24 is less than the energy of the return signal y {n) 14, and the energy of the error signal (n) is less than the energy of the feedback signal y (n) 14; 3. When the source of the echo has disappeared, and the system has converged:

in an echo-only situation, the signal echo (n) is zero, the signal y (n) 14 is zero, the estimated signal (n) is strong, and the error signal (n) is strong;

in the situation of local speech alone, the error signal (n) is close to the signal y (n) 14, and the estimated signal (n) 24 is close to zero;

in a double-talk situation, the energy of the error signal (n) is greater than the energy of the signal y (n). According to the coupling value between the return channel and the direct channel, and the level of the signal. signal y (n) 14, the energy of the estimated signal (n) 24 is either of the order of magnitude, or greater or less, than the energy of the signal y (n) 14.

In other words, before convergence of the system, that is to say before adaptation of the coefficients of the adaptive filter 16, the energy of the return signal 14 y (n) carried by the return path is always significantly greater than the energy the estimated estimated echo signal (n) 24, whether in an echo-only, local-only, or double-talk situation.

After convergence of the coefficients of the adaptive filter, making it possible to eliminate, or at least reduce, the echo components of the direct signal x (n) 11 present in the feedback signal y (n) 14, the signal energy y (n) 14 is substantially equal to that of the estimated signal (n) 24 in an echo situation alone, the energy of the signal y (n) 14 is much greater than that of the estimated signal (n) 24 in the speech situation local only, and the energy of the signal y {n) 14 is slightly greater than that of the estimated signal (n) 24 in double talk situation.

After suppressing the echo source and convergence of the coefficients of the adaptive filter, the energy of the signal y (n) 14 is much lower than that of the estimated signal (n) 24 echo only, and the energy of the signal y (n) 14 is significantly higher than that of the estimated signal (n) 24 in situation of local speech alone.

Once these different equations are set, it is found that the energy of the signal y {n) 14 is always greater than or substantially equal to that of the estimated signal (n) 24, except in the case where the echo source disappears. Indeed, in a situation of echo path change, and in particular of echo cancellation, the energy of the signal y (n) 14 is much lower than that of the estimated signal (n). This situation appears systematically in echo situation alone and may appear in a double talk situation, depending on the value of the coupling.

It is therefore possible to detect a change in the echo path by comparing a first piece of information representative of the energy of the estimated signal (n) and a second piece of information representative of the energy of the signal y (n). the echo cancellation technique according to the invention is based on a comparison of the ratio R of the energies of the signals estimated (n) and y (n) to a first threshold S1, a change in the echo path being detected when this ratio R exceeds the threshold Sl: energy (estimated (n)) energy (}> (») When an echo path change has taken place, the ratio R is greater than the threshold S1 only in echo situation alone.

It can notably be noted that it is necessary to detect this situation very rapidly, in order to prevent the adaptive filter 16 from acting as a noise generator, while avoiding false alarms, which occur when the estimated signal energy (n ) exceeds that of the signal y {n) for a very short time.

To do this, the ratio R is compared with the threshold Sl, and when the value of the ratio R exceeds this threshold Sl, timing means, for example comprising a counter, are triggered. The decision to detect the echo cancellation is taken once the counter exceeds a second maximum value threshold 52, in order to avoid false false detections.

The following is an example of a preferred embodiment of the invention, as illustrated in FIG.

During a first step, an estimate of the energy of the estimated estimated echo signal (n) 24 at the output of the adaptive filter 16 is determined from first means 21 for determining a first piece of information representative of the energy of the estimated echo signal. An estimation of the energy of the feedback signal y (n) 14 at the input of the system is also determined from second means 22 for determining a second piece of information representative of the energy of the return signal.

The energy estimate of the echo and return signals can for example be determined using the following equation: energy (X) = a * energy (X) + (l-a) * X ² with a = 0.996.

In a next step, comparison means 23 determine the ratio R between the estimate of the energy of the output of the adaptive filter 16 and the estimate of the energy of the feedback signal. By calculating the estimated energies of the estimated echo signal and the return signal, and by making the ratio R of these energies, it is thus possible to detect the change of physical coupling of the system.

The ratio R is then compared with the threshold S1, and a counter is triggered if

R is greater than Sl:

if R> Sl, then we increment the value "tempo" of the counter (Tempo ++).

Finally, we compare the value "tempo" of the counter with a threshold S2 corresponding to a maximum value of time delay, and if the value "tempo" is above this threshold 52, it is decided that an echo cancellation has been detected.

The choice of the thresholds Sl and 52 makes it possible in particular to adapt the detection technique to the system to which it applies. These thresholds may for example be predetermined or adapted as a function of control information representative of the state of the device. Thus, the thresholds S1 and / or 52 can be fixed "absolute", that is to say fixed and not changing over time. The thresholds S1 and / or 52 may also be predetermined as a function of control information representative of the state of the device, taking into account, for example, the nominal value of the coupling (in other words the level of the echo return) for the threshold. Sl, or the current value of the overflow delay and / or the number of false alarms for the thresholds S1 and / or 52. Finally, these thresholds S1 and / or 52 can also be adapted over time.

For example, we can choose Sl such that: Sl = β * CouplingValue, where β is a constant.

Thus, the larger the thresholds S11 and 52, the more reliable the detection of a change in the echo path, but the sooner the detection is to be active.

For echo cancellers, a compromise is to be determined to avoid the unpleasant generation of an echo at the output of the system. Once set, the system allows a sufficiently fast and reliable trigger not to perceive the echo cancellation.

According to an advantageous embodiment of the invention, given as a simple illustrative and non-limiting example for a given application, one can choose Sl equal to 128, and 52 equal to 25.

Once the suppression or echo path change has been detected, the control means 26 (comprising the first means 21 and the second means 22 for determining an information representative of the energy, and the comparison means 23) act on the means of treatment, to put the system in a determined state. Thus, the control means 26 may for example reset the coefficients of the adaptive filter 16, or cancel the call to an echo cancellation function, in order to stabilize the state of the system.

The invention thus proposes a technique for canceling an acoustic or electrical echo, based on the use of control means of the echo processing means, as a function of the detection of change of physical echo path. More specifically, the invention proposes detecting an echo suppression by comparing the energy estimates of an estimated echo signal at the output of the processing means and a feedback signal entering the system. .

This technique has the advantage of dispensing with means for detecting the state of the system, as proposed according to the prior art. According to the invention, there is therefore no interaction with a system status indicator for the detection of the change of the echo path, which makes it possible to increase the reliability of an echo cancellation device according to the invention.

The invention finds applications in the acoustic and / or electrical echo cancellation devices, in particular for integration into terminals enabling the parallel communication of several telephone lines, such as terminals dedicated to the stock markets, for example.

Claims

An echo canceling device present in a return path (14), and corresponding to a direct path (11) of an audio communication system, comprising echo processing means comprising filtering means adaptive device (16) of said direct channel, said echo processing means delivering an estimate of said echo to be subtracted from said return channel (14), characterized in that it comprises means (26) for controlling said means processing method taking into account a comparison between a first information representative of the energy output of said processing means and a second information representative of the energy of said return path (14).

2. An echo cancellation device according to claim 1, characterized in that said control means determines a ratio between said first and second information and that said comparison compares said ratio to a first threshold.

3. An echo cancellation device according to claim 2, characterized in that said control means comprise triggering means triggered when said ratio is greater than said first threshold.

4. echo cancellation device according to claim 3, characterized in that said delay means comprise at least one counter.

An echo cancellation device according to any of claims 3 and 4, characterized in that said control means outputs echo path change information when said delay means exceeds a second threshold.

6. An echo cancellation device according to any one of claims 1 to 5, characterized in that said control means allow to act on at least one coefficient of said filtering means.

7. An echo cancellation device according to any one of claims 1 to 6, characterized in that said control means allow the inhibition of said filtering means.

8. An echo cancellation device according to any one of Claims 2 to 7, characterized in that said first threshold is predetermined or adapted according to a first control information representative of a state of said device.

9. An echo cancellation device according to any one of claims 5 to 8, characterized in that said second threshold is predetermined or adapted according to a second control information representative of a state of said device.

A method of canceling an echo present in a return path (14), and corresponding to a direct path (11) of an audio communication system, comprising an echo processing step implementing a step of adaptive filtering said direct channel, said echo processing step delivering an estimate of said echo to be subtracted from said return channel (14), characterized in that it comprises a step of controlling said step of treatment taking into account a comparison between a first information representative of the energy output of said processing step and a second information representative of the energy of said return path (14).

11. Computer program product downloadable from a communication network and / or stored on a computer readable medium and / or executable by a microprocessor, characterized in that it comprises program code instructions for the implementation of the method of canceling an echo according to claim 10.