FR2785481A1 - Hands free telephone sound echo reduction system having separate position microphones with outputs amplitude and phase measured and echo reduction application unit/loudspeaker applied. - Google Patents

Abstract

The echo reduction system has a loudspeaker (15) connected to an output application unit (14). The sound detector has two units (11,12) spaced apart, and the outputs are passed to a processing unit (13) which detects amplitude and phase of the sound signals. the signals are applied to the application unit to provide a reduced echo sound output.

Description

The present invention relates to a sound capture and restitution system

  provided with means for reducing the echo resulting from the acoustic coupling between its capture means and its restitution means. Such a system is for example, but not necessarily, a telephone set intended to operate in a hands-free position. Sound capture and reproduction systems generally consist of a sound capture means such as a microphone, a sound reproduction means, such as a speaker, and an application unit , such as an amplifier which will increase its power and then transmit it by means of restitution or such as a telephone unit which, on the one hand, will send it to other telephone units and,

  on the other hand, after amplification, will transmit it by means of restitution.

  The signal emitted by the restitution means can be picked up by the pickup means and this results in an acoustic echo. This acoustic echo is broken down into two essential parts: the direct speaker / microphone coupling and the response of the

room.

  If the response of the room is generally effectively treated by conventional echo cancellation techniques (due, among other things, to the fairly low level of this part of the echo), it is not the same for coupling. direct, which results from both the acoustic mode and the propagation of vibrations, for example through the housing if the system in question is a telephone terminal. This part of echo due to direct coupling generally has a higher power than the local speech picked up by the pickup means (from 6 to 18 dB, i.e. a multiplier coefficient of 8), and leads to a phenomenon of auto oscillation generally called "Larsen effect" if none

  adequate treatment is not applied.

  If echo cancellation makes it possible to deal more or less with this harmful effect, a substantial gain variation must be implemented in addition to avoid departure in feedback during the convergence phase of the echo canceller, but also thereafter for prevent any problem if the echo path varies or during moments of double talk. For a telephone terminal, direct acoustic coupling is important since the microphone is placed in the same case as the loudspeaker. The echo that results from this coupling is the major obstacle to the proper functioning of stations

hands-free phone calls.

  We therefore sought to reduce the effects of direct coupling. The micro distance

  speaker cannot be changed in significant proportions given the size of the individual terminals. It should be noted, however, that each time the

  distance is doubled, the power of the echo is divided by four.

  It is however possible to intervene on the mechanics of the coupling between the speaker and the microphone. For example, the microphone is separated from the housing using anti-vibration materials, such as foam or rubber, which has

  the effect of suppressing or greatly reducing the vibrations generated by the built-in

  speaker and transmitted, via the box, to the microphone.

  This solution of mechanical decoupling effectively reduces coupling by vibration, but turns out to be an industrially expensive solution. In addition, it does not reduce the acoustic coupling which can turn out to be important if the microphone is placed relatively close to the loudspeaker, which is today the trend because of the

  mode for small terminals.

  Another known solution would consist in using, in parallel with the path of the echo signal, a compensation filter whose response is inverse to the impulse response of the coupling between the microphone and the loudspeaker, both mechanical and acoustic coupling. In theory, at the output of the compensation filter, the signal from the speaker due to the echo is canceled. However, in practice, this technique is quite advanced and is no longer satisfactory if the coupling has changed in its characteristics, even in a minimal way, following for example a disassembly / reassembly of the terminal. This technique is also not suitable if the nature of the disturbances is not linear, that is to say if they cannot be modeled by the product of the signal coming from the loudspeaker and of a linear function which more is, the operation of transducers is rarely linear, as they are generally subject to distortions and / or saturations, which are typical examples

nonlinear functions.

  Finally, for mass production, the coupling will necessarily be different from one terminal to another, due for example to the use of slightly different speakers or microphones, again making this technique of compensation by

ineffective filter.

  The object of the present invention is therefore to propose an acoustic system intended to reduce the echo which results from direct acoustic coupling between its means of sound capture and its restitution means which makes it possible to solve the problems.

mentioned above.

  To this end, a sound pickup and playback system according to the invention is characterized in that said sound pickup means consists of at least two sound sensors located at different distances from said pickup means and a processing unit designed to exploit the amplitude and phase of the signals from each sound sensor in order to deliver a signal to said application unit, the

echo signal is reduced.

  According to another characteristic of the invention, said processing unit operates a combination of the signals respectively from the sound sensors, the signal from a sensor being delayed by a duration corresponding to the propagation time of said signal

  echo from said rendering means between said sensor and a reference sensor.

  According to another characteristic of the invention, said combination operated by said processing unit also consists in applying a multiplying coefficient from the signal from each sensor equal to the inverse of the attenuation of the echo signal from

  said restitution means with respect to a reference sensor.

  According to another characteristic of the invention, said processing unit performs a filtering of the signal from each sensor and then a summation of the signals thus filtered, the impulse response of each filtering associated with a sensor being defined by a function g, (t) which is a solution of the following equation: - [h, (t) * g, (t)] O 0 for i varying from I to n //,..,-, coh, (t) (i varying from 1 to n) is the impulse response of each sensor to

  signals delivered by said restitution means.

  According to another characteristic of the invention, one of the filtering is defined by a

multiplication function.

  According to another characteristic of the invention, the impulse response of each sensor to the signals delivered by said rendering means is determined by

  measuring the response of said sensor to signals emitted by said rendering means.

  According to another characteristic of the invention, each function g, (t) is determined by solving said equation by means of an adaptive algorithm. According to another characteristic of the invention, said system comprises two

  sensors, one of which is closer to the restitution means than the other.

  According to another characteristic of the invention, the filtering associated with the most

  distant from the restitution means is defined by a multiplication function.

  The features of the invention mentioned above, as well as others,

  will appear more clearly on reading the following description of an example of

  realization, said description being made in relation to the attached drawings, among

  which: Fig. 1 is a block diagram of a collection means of a system according to the invention, FIG. 2 is another block diagram of a capture means according to the invention, in a first embodiment of the filtering function, FIG. 3 is another block diagram of a sound capture and restitution system according to the invention, in a second embodiment of the filtering function, and FIG. 4 is a diagram illustrating the efficiency of a collection system and

  sound reproduction according to the invention.

  In Fig. 1, a sound capture and restitution system 100 has been shown, the sound capture means 10 of which consist of two sound sensors 11 and 12, here made up of microphones, respectively connected to two inputs of a processing unit 13 The unit 13 has its output which is connected to the input of an application unit 14, for example an amplifier or a telephone unit. The system 100 also includes sound reproduction means 15, here a loudspeaker, connected to the output of said telephone unit 14. Also shown in FIG. 1, a sound source S.

  The microphone 11 is the main microphone which receives the useful signal denoted s (t).

  The sound signals emitted by the source S are picked up by the microphones 11 and 12 then delivering electrical signals which are then processed in the unit 13 before being transmitted to the application unit 14. The latter is for example a amplifier which will increase the power of the signals received from the unit 13 and deliver them to the loudspeaker 15 for reproduction purposes. It may also be a telephone unit which, on the one hand, will send these signals by telephone to one or more other subscribers and, on the other hand, will increase their power and deliver them to the

  loudspeaker 15 for restitution purposes.

  According to a characteristic of the invention, the two microphones 11 and 12 are at different distances from the loudspeaker 15, the microphone 12 being closer than the microphone 11. As will be seen later, the other geometric specifications, such as symmetry of revolution, are not relevant to the invention. According to the invention, this is the phase and amplitude information of the signals

  from microphones 11 and 12 which are processed in the processing unit 13.

  According to a first embodiment of the invention, this processing corresponds to a combination of these signals with delay from the second coming from the microphone 12 closest to the loudspeaker 15 relative to the first coming from the main microphone 11. This delay, noted te, corresponds to the propagation time between the two microphones 11 and 12 of the echo signal from the loudspeaker 15. It is shown below that such a combination makes it possible to reduce the acoustic echo resulting from the coupling between the

  main microphone 11 and speaker 15.

  We will consider the signal x (t) at the output of the main microphone 11, the one that is furthest from the speaker 15, x2 (t) the signal at the output of the microphone 12, the one that is therefore the closest, and y (t) the signal at the output of the processing unit 13 after

  processing of the two signals x, (t) and x2 (t).

  The processing unit 13 operates a processing of the two signals x, (t) and x2 (t2) respectively from the microphones 11 and 12 consisting in applying to the signal from the microphone 12 a multiplying coefficient equal to the inverse of its attenuation at compared to that coming from micorphone 11 and a delay equal to the delay of the echo signal t ,, compared to micorophone 11 and summing with the signal coming from microphone 11. We thus have: y (t) = xt) + - at. X2 (t - tO e We will also consider, for the sake of simplification of the model, that the propagation is in free field. If in addition, we neglect the background noise in the room where said terminal 10 is located as well as the noise microphones 11 and 12 and loudspeaker 15, we can write the following relation with regard to microphone 11: x (t) = s (t + e (t) os (t) and the signal from the sound source S (s (ft) also corresponds to

  useful signal) and e (t) is the so-called echo signal from speaker 15.

  As regards the microphone 12, the signal from the source S is, compared to that received by the microphone 11, attenuated by a coefficient oh and delayed by a time ts. As for the echo signal from speaker 15, it is attenuated by a coefficient

  aoh and in phase advance with respect to the signal received by the microphone 11 of the time t <.

  We can therefore write: x2 (t) = a, s (t-ts) + ae e (t + t) For the purposes of the demonstration, we will assume that the source S is relatively far from microphones 11 and 12. It it follows that the amplitudes of the signal

  useful on the two microphones 1 1 and 12 are substantially identical (a. _- 1).

  It will be recalled that the microphone 12 is advantageously closer to the loudspeaker.

  speaker 15 than microphone 11. As a result, the amplitude of the echo signal on microphone 12 is greater than that on microphone 11 and the coefficient o4 is

greater than 1.

  The signal at the output of the processing unit 13 can therefore be written: 1 1 y (t) = X = s (t) - s (t - ts - t) a ,, Because the coefficient ae is higher at one, the signal at the output of unit 13

is close to the useful signal s (t).

  It will also be noted that it would be the same if the delay difference te on the microphones 11 and 12 of the echo signal is substantially equal to the delay difference 6 on the microphones 11 and 12 of the source S. We now describe a second embodiment of a sound capture means 10 according to the invention in relation to FIG. 2 which is a block diagram comprising the same elements as those of the system shown in FIG. 1. Each

  microphone 11, 12 is connected to the input of an analog / digital converter 16, 17.

  The digital signal from the converter 17 is delivered to the input of a filter 18 whose impulse response is described by a function g (t). The signals from, on the one hand, the converter 16 and, on the other hand, the filter 18 are delivered on the input of an adder 19 whose output is connected to the input of a digital / analog converter 20 The signal at the output of converter 20 is delivered to the unit

7 2785481

  of application 14. It will be noted that the unit 13 incorporates here the functions of the filter 18 thus

than that of the summator 19.

  It will be understood that if microphones 11 and 12 are available which deliver digital signals directly or that, conversely, if there is a filter 18 with analog operation, the converters 16, 17 and 20 are not not necessary. We will describe the operation of this embodiment. But beforehand, we will call h / (t) and h2 (t) the impulse responses of microphones 11 and 12 respectively concerning the signal from loudspeaker 15. The filter 18 performs a processing consisting in applying only to only signal from microphone 12 processing which can be represented by the function g (t). Furthermore, the summator 19 delivers a signal y (t) which can be written as follows: y (t) = xi (t) + g (t) * x2 (t) Equation 1 We could show that l the echo is considerably reduced if the function g (t) which defines the processing performed by the filter 18 is such that it is a solution of the following equation: hi (t) + g (t) * h2 (t) = O Equation 2 The signal delivered by the microphone 11 can be written in the form: x, (t) = s (t) + h, (t) * e (t) Equation 3 o as previously s (t) is the signal received from the source S (also useful signal), e (t) is the signal received from the loudspeaker 15 and the sign * represents the convolution operation. As for the signal delivered by the microphone 12, it can be written: x2 (t) = u (t) * s (t) + h2 (t) * e (t) Equation 4 where (t) is the impulse response of the microphone 12 as regards the signal coming from the source S. 8s 2785481 It will be noted that the mathematical modeling is more realistic than the preceding one insofar as the latter was made by supposing that the echo signal e (t) was only delayed and weakened, which would have been the case in a context of free propagation with a loudspeaker 15 having ideal characteristics (omnidirectional, etc.). These assumptions are not made here. It can be shown that, given equations 1 to 4 above, the signal at the output of the processing unit 13 can be written: y (t) t [t) + g (t) * ut)] * s (t) Equation 5

  o "t) represents the impulse of Dirac.

  We can see that, according to this equation, the echo is canceled. More particularly, the useful signal s (t) is little distorted if g (t) * u (t) is small compared to the pulse of Dirac "t). This is all the more the case if the echo is more loud on microphone 12 than on microphone 11, especially if microphone 12 is more

  close to the speaker 15 than the microphone I l.

  To solve equation 2 above in order to determine the function g (t), one can either use non-adaptive techniques such as the so-called least squares technique, or use an adaptive algorithm for a treatment which adapts during the

time.

  However, it has been shown that a fixed treatment can be sufficient in the majority of applications, which has the advantage of being an economical solution. We can therefore proceed as follows: 1) impulse response measurements of the microphones 11 and 12 are carried out on a signal emitted by the loudspeaker 15, and this in the geometric configuration of the terminal 10. The purpose of this operation is to subsequently calculate the responses

impulse h, (t) and h2 (t).

  2) the above equation is then solved by a deconvolution method in order to

  determine the function g (t), which we implement in unit 14.

  There is shown in FIG. 3 another embodiment of a capture means of a capture and restitution system according to the present invention. In this embodiment, the device 30 includes n microphones 311 to 31, respectively connected to analog / digital converters 32, to 32 ,. The respective outputs of

9 2785481

  converters 321 to 32 ,, are respectively connected to the inputs of filters 33, to 33, the outputs of which are connected to inputs of an adder 34 itself connected to the input of a digital / analog converter 35. The output of this converter 35 is connected to the input of a telephone unit 14 (identical to that of the first embodiment), one output of which is connected to a loudspeaker 15. The microphones 311 to 31 ,, are at different distances of the speaker 15 and, advantageously, to take account of the fact that the speaker 15 is not necessarily ideal, in particular omnidirectional, are placed around the speaker 15. It will be understood that if digital microphones are available or, conversely, if analog filters are available, the converters 322 to 32 ,, and 35 are not

not necessary.

  The impulse response to the signals delivered by the loudspeaker 36 of each microphone 3 1 i is h, (t) (i varying from 1 to n) and is either predetermined by the geometry (delay, attenuation), or measured once for all, either estimated by an adaptive algorithm If we call g, (t) the function corresponding to the impulse response of each filter 33j (i varying from 1 to n), this must be a solution of the following equation: _ [h, (t) * g, (t)] t 0 Equation 6 According to a preferred embodiment, like the first embodiment shown in FIG. 2, one of the filters, for example the filter 331, is of length I so that it reduces to a multiplication. This saves a

  filter, therefore reducing the computational load.

  In a simplified embodiment with several sensors, said processing unit 13 operates a combination of the signals respectively from the sound sensors, each signal from a sensor being advanced by a duration corresponding to the propagation time of said echo signal from said rendering means between said sensor and a reference sensor, which can be chosen arbitrarily, for example the sensor 31 l

in Fig. 3.

  The curve of FIG. 4 shows, in the case of a device with two microphones such as that which is shown in FIG. 1, the impulse response of the microphone 11 (upper curve) and the total impulse response at the end of the treatment operated by

2785481

  the processing unit 13 (lower curve) as a function of time, in the case of taking a hands-free sound from a system 100 with two microphones 11 and 12 having the

above features.

  We therefore note that the effect of the reduction in coupling relates to the first part of the impulse response, which corresponds to the direct loudspeaker - microphones path. Note that, in the case of non-adaptive treatment, it is almost impossible

  to correct the whole response whose tail varies over time.

  In the embodiment which led to the drawing of the curves of FIG. 4, the associated filter

  at microphone 12 had 100 coefficients.

Claims (8)

  1) Sound capture and restitution system of the type comprising sound capture means and sound restitution means (15) connected to an output of an application unit (14) to which said signal from said signal is applied means of capture for the purposes in particular of restitution by said restitution means (15), said system being provided for reducing the echo resulting from the acoustic coupling between said sound capture means and said restitution means, characterized in that said means sound collection system consists of at least two sound sensors (11, 12, 311 to 3 l1) located at different distances from said restitution means (15) and from a processing unit (13) designed to operate the amplitude and phase of the signals from each sound sensor (I 1, 12, 33, at 33n) in order to deliver a signal to said application unit (14) whose the echo signal is reduced.   2) Sound collection and restitution system according to claim 1, characterized in that said processing unit (13) operates a combination of the signals respectively from the sound sensors (11, 12, 31l to 31 ,,), each signal coming from a sensor being delayed by a duration corresponding to the propagation time (te) of said echo signal coming from said reproduction means (15) between said sensor and a reference (I 1, 311).   3) System for capturing and restoring sound according to claim 2, characterized in that said combination operated by said processing unit (13) consists in addition of applying a multiplying coefficient derived from the signal from each sensor equal to the inverse of the attenuation of the echo signal from said reproduction means (15) by   compared to a reference sensor.   4) Sound capture and restitution system according to claim 1, characterized in that said processing unit (13) performs a filtering of the signal from each sensor (11, 12, 31 to 3 1n) then a summation of the signals thus filtered, the impulse response of each filtering associated with a sensor (I l, 12, 31, to 31,) being defined by a function g, (t) which is a solution of the following equation Z [h, (t) * g, (t)] z 0 for i varying from I to noh, (t) (i varying from I to n) is the impulse response of each sensor (I l,   12, 3 1 to 3 1n) to the signals delivered by said restitution means (15). 12 2785481   ) Sound capture and restitution system according to claim 4, characterized   in that one of the filtering is defined by a multiplication function.   6) Sound collection and restitution system according to claim 4 or 5, characterized in that the impulse response of each sensor (11, 12, 31 to 31,) to the signals delivered by said restitution means (15) is determined by measuring the response of said sensor (11, 12, 311 to 31,>) to signals emitted by said means of restitution (15).   7) Sound capture and restitution system according to one of claims 4 to 6,   characterized in that each function g, (t) is determined by solving said equation by means of an adaptive algorithm   8) Sound capture and restitution system according to one of claims   above, characterized in that it comprises two sensors (11 and 12), one of which is more   closer to the restitution means (15) than the other.   9) Sound capture and restitution system according to claim 8, characterized in that the filtering associated with the sensor (11) furthest from the restitution means   (15) is defined by a multiplication function.