EP0884926B1 - Method and device for optimized processing of an interfering signal when recording sound - Google Patents

Description

The invention relates to a method and a device for optimized processing of a disturbing signal during a sound recording.

With the joint advent of the era of trade information, audio and / or video information, technicians of research and development of means of access to this information are most often confronted, in the most areas of application and use of these information, to the general problem of estimating a useful signal, bearer of this information, from one or of several observation signals, composed of this signal useful degraded due to the presence of disturbing signals.

In the more specific area of sound recording, these signals corresponding to audio frequency signals, this problem is most often resolved by putting it into operation concomitant, co-operation, of several devices processing of this observation signal, each of these devices being optimized locally so as to reduce, significantly, at one of these devices, the influence of a particular component of these signals disruptive or at least one of these interfering signals.

Under these conditions, problems appear interaction between these different devices, which, well heard, makes it difficult to optimize the different applied treatments. Modification, for optimization, control parameters for a particular device generally requires mutual modification of those of other devices used.

In addition, the co-operation of these different devices leads to a complexity of implementation not optimized and generally at high cost.

Various examples of conventional solution known from the state of the art will be given below in conjunction with FIGS. 1a to 1d. In general, the observation signal y (t) can be considered as the sum of the original useful signal s (t) and a disturbing signal p (t) according to the relation: y (t) = s (t) + p (t). The disturbing signal itself can be considered as the sum of N elementary components and verifying the relation:

As illustrated in FIG. 1a, a common solution proposed for solving such a problem may consist in co-operating a number N of devices, each of them being optimized and dedicated to the reduction or even the local elimination of 'a given component p _k (t) of the disturbing signal.

Such an approach leads to successively minimizing a local estimation error linked to each component of the disturbing signal. Each of these successive minimizations thus amounts to locally implementing a processing T _k (t) adapted to the component p _k (t) of the corresponding disturbing signal.

The general principle of treatment, known as as shown in figure la, is in particular used when taking his hands free in the context of mobile radio as well as in the context of the videoconferencing.

In the context of radiotelephone applications hands-free for mobiles, the disturbing signal p (t) can be considered as composed of an observation noise b (t), vehicle rolling noise, aerodynamic noise such as wind, air flow, as well as an acoustic echo signal z (t) from the coupling acoustic between speaker and plug microphone of his.

• B.AYAD, G.FAUCON et R.LE BOUQUIN JEANNES, "Optimization of a Noise reduction preprocessing in an acoustic echo and noise controller", IEEE International Conference on Acoustics, Speech, and Signal Processing Conference, pp. 953-956, Atlanta, USA, May 7-10, 1996 ;
• Y.GUELOU, A.BENAMAR et P.SCALART, "Analysis of two structures for combined acoustic echo cancellation and noise reduction", IEEE International Conference on Acoustics, Speech, and Signal Processing Conference, pp. 637-640, Atlanta, USA, May 7-10, 1996 ;
• R.MARTIN, P.VARY, "Combined acoustic echo control and noise reduction for hands-free telephony - State of the Art and perspectives", proceedings of the Eighth European Signal Procesing Conference, pp. 1127-1130, Trieste, Italy, 10-13 September, 1996.

In order to minimize the influence of these two components of the disturbing signal and to transmit a better quality signal to the distant correspondent, current research and work has proposed cascading a noise reduction system and a acoustic echo control system. Such a combination of systems is shown in Figure 1b. The general principle of the solutions thus proposed consists in placing a filter noise reduction device RB downstream, as shown in FIG. 1b, or upstream of the acoustic cancellation device, filter H _t . For a more detailed description of this type of device, one can usefully refer to the most recent articles, published by:

• B.AYAD, G.FAUCON and R.LE BOUQUIN JEANNES, " Optimization of a Noise reduction preprocessing in an acoustic echo and noise controller ", IEEE International Conference on Acoustics, Speech, and Signal Processing Conference, pp. 953-956, Atlanta, USA, May 7-10, 1996;
• Y. GUUELOU, A. BENAMAR and P. SCALART, "Analysis of two structures for combined acoustic echo cancellation and noise reduction ", IEEE International Conference on Acoustics, Speech, and Signal Processing Conference, pp. 637-640, Atlanta, USA, May 7-10, 1996;
• R.MARTIN, P.VARY, " Combined acoustic echo control and noise reduction for hands-free telephony - State of the Art and perspecti ves", proceedings of the Eighth European Signal Procesing Conference, pp. 1127-1130, Trieste, Italy, 10-13 September, 1996.

In the context of videoconferencing applications, the disturbing signal p (t) can be considered as composed, not only of an observation noise b (t) and a acoustic echo signal z (t), but also of a signal r (t) generated by the reverberation effect of the room where sound recording is performed.

The solutions proposed, in such a context, can be classified into two main types, depending on whether the signal is considered to be essentially harmful echo and noise, or noise and reverberation.

In the two aforementioned cases, the solutions adopted correspond to a cascade of elementary treatments, each of them being adapted to a particular component of the disturbing signal.

• R.MARTIN et P.VARY
  "Combined acoustic echo cancellation, dereverberation and noise reduction : a two microphone approach",
  Annales des télécommunications, Tome 49, n° 7-8, pp. 429-438, 1994.

According to the first type of these solutions, as shown in FIG. 1c, two elementary treatments are implemented: an echo cancellation processing and a processing whose object is to reduce the influence of noise, RB filter, on the useful signal.
In the more specific case of FIG. 1c, in which there are also two microphones for implementing the sound pickup system, a copy of the RB filter is applied to the signal broadcast on the loudspeaker, in order to reduce the influence non-linear variations of this filter on the process of identifying the echo signal. For a more detailed description of the noise and echo processing, one can usefully refer to the article published by:

• R.MARTIN and P.VARY
  " Combined acoustic echo cancellation, dereverberation and noise reduction: a two microphone approach ",
  Annales des telecommunications, Tome 49, n ° 7-8, pp. 429-438, 1994.

• C.MARRO, Y.MAHIEUX et K.U.SIMMER,
  "Performance on adaptive dereverberation techniques using directivity controlled arrays", Proceedings of the Eighth European Signal Processing Conference, pp. 1127-1130, Trieste, Italy, 10-13 September, 1996 ;
• K.U.SIMMER, S.FISHER et A.WASILJEFF,
  "Suppression of coherent and incoherent noise using a microphone array", Annales des télécommunications, Tome 49, n° 7-8, pp. 439-446, 1994.

According to the second type of these solutions, as shown in FIG. 1d, sound recording can be carried out from a large number of microphones so as to produce an acoustic antenna having the object of focusing the main lobe of the antenna on the speaker and thus privilege the area of space where the speaker is actually located to carry out an operation of noise reduction and reverberation. The acoustic antenna conventionally includes a number of band filters F ₁ to F _N and a summator, performing antenna processing. Another post-filtering treatment is applied at the antenna output and consists in reducing the remaining reverberation. For a more detailed description of this type of solution, one can usefully refer to the articles published by:

• C. MARRO, Y. MAHIEUX and KUSIMMER,
  " Performance on adaptive dereverberation techniques using directivity controlled arrays ", Proceedings of the Eighth European Signal Processing Conference, pp. 1127-1130, Trieste, Italy, 10-13 September, 1996;
• KUSIMMER, S.FISHER and A.WASILJEFF,
  " Suppression of coherent and incoherent noise using a microphone array ", Annals of telecommunications, Tome 49, n ° 7-8, pp. 439-446, 1994.

In all the above-mentioned solutions, the cascading of these elementary processing, each of them being adapted to only one of the components of the disturbing signal, leads to obtaining a sub-optimal solution to the general problem of rejection of the disturbing signal and induces, in addition, a significant production cost.
Indeed, each of these processing operations minimizing a local error, because relating to an elementary or local component of the disturbing signal, their association generally does not lead to the global minimum of the optimal solution.

In addition, the practical implementation of each of these elementary treatments are only an approximation ideal processing, signal distortions useful being introduced for each treatment, from the point of view of other treatments, which ultimately can lead to an input of the useful signal transmitted strongly degraded compared to the original useful signal.

Finally, cascading these treatments elementary requires studying the optimal position and the interaction of different elementary treatments, relative to each other, in order to get the best configuration. It should however be noted that the conclusions of such a study should be questioned depending on the choice of processes and algorithms used to conduct the various elementary treatments. Such a constraint is described in the article published by Y.GUELOU, A.BENAMAR and P.SCALART, 1996, previously cited, in the case hands-free mobile telephony. Parameterization, in view of their adjustment, of the processes and algorithms implemented then appears delicate, the modification of a given parameter generally requiring a consequential modification of at least minus certain parameters of the other elementary treatments.

EP-A-0 767 569 describes an echo processing system acoustics by an adaptive filtering technique. A special feature of this system lies in the use of filter control parameters, not adaptation and forgetting factor, variable over time and adapting automatically to the acoustic environment.

A posteriori optimization of these treatments may, where appropriate, be considered. Such a procedure inevitably implies, on the one hand, a permanent exchange information between these elementary treatments and, on the other hand, the application of concerted constraints on the adjustment parameters of these. Such a posterior optimization of such systems has shown the limitations of this approach because of the results ultimately obtained.

The object of the present invention is to remedy the shortcomings and disadvantages of processes, processes and prior art systems previously described.

Such an object is achieved by the implementation of a a priori optimization process of signal processing disruptive affecting any observation signal, this being completely separate, i.e. the prior art described previously in the description, or even any posterior optimization of the processes supra.

The process of a priori optimization of treatment a disturbing signal during a sound recording, from an observation signal formed by an original useful signal and this disturbing signal is implemented thanks to a method and device consisting of, respectively allowing an estimation of the disturbing signal to generate an estimated disturbing signal. An estimation useful signal to generate an estimated useful signal and a filtering of the observation signal from the disturbing signal estimated and optimal filtering allow minimization of the error between the useful signal and the estimated useful signal. The estimated useful signal converges to the original useful signal for a substantially zero error between the useful signal and the estimated useful signal.

The method and the device, objects of the invention, find application in any context relating to the taking of sound, including hands-free mobile telephony, hands-free video conferencing, and more generally studio or audio operations.

• la figure 2a représente, à titre d'exemple non limitatif, un schéma bloc illustratif de la mise en oeuvre du procédé, objet de la présente invention, dans le domaine temporel ;
• la figure 2b représente, à titre d'exemple non limitatif, un schéma bloc illustratif de la mise en oeuvre du procédé, objet de la présente invention, dans le domaine temporel, dans le cas plus particulier de l'existence d'un signal de réception générateur d'un signal d'écho apportant une contribution spécifique au signal perturbateur ;
• la figure 2c représente, à titre d'exemple non limitatif, dans une situation semblable à celle de la figure 2a, un schéma bloc illustratif de la mise en oeuvre du procédé, objet de la présente invention, dans le domaine fréquentiel ;
• la figure 2d représente, à titre d'exemple non limitatif, un schéma bloc illustratif de la mise en oeuvre du procédé, objet de la présente invention, dans une situation semblable à celle de la figure 2b, dans le domaine fréquentiel, dans le cas particulier d'un signal de réception générateur d'un signal d'écho apportant une contribution spécifique au signal perturbateur ;
• la figure 2e représente, à titre d'exemple non limitatif, un schéma bloc illustratif d'une mise en oeuvre préférentielle par un traitement par blocs successifs du signal d'observation, dans une situation semblable à celle de la figure 2d, dans le cas de l'existence d'un signal de réception générateur d'un signal d'écho apportant une contribution spécifique au signal perturbateur ;
• la figure 3a représente, sous forme de schémas blocs, le schéma synoptique d'un dispositif permettant, dans le domaine fréquentiel, le traitement général, respectivement le traitement par blocs successifs, du signal d'observation, dans le cas général de l'existence d'un signal de réception, générateur d'un signal d'écho apportant une contribution spécifique au signal perturbateur ;
• la figure 3b représente un détail de réalisation avantageux d'un module d'estimation de la densité spectrale de puissance du signal utile plus particulièrement mis en oeuvre dans le dispositif représenté en figure 3a, lorsque, en particulier, le traitement par bloc est mis en oeuvre ;
• la figure 3c représente une variante de réalisation du dispositif représenté en figures 3a ou 3b, dans laquelle un module d'estimation de la densité spectrale de l'écho d'un signal de réception et un module d'estimation de la densité spectrale du signal de bruit, dans le contexte d'une application à la radiotéléphonie mobile mains libres, sont introduits ;
• les figures 3d et 3e représentent, à titre d'exemple non limitatif, un module d'estimation de la densité spectrale de puissance du signal de bruit et du signal d'observation, par filtrage récursif à partir d'un facteur d'oubli ;
• les figures 4a à 4e représentent différents chronogrammes de signaux, relevés en des points de test remarquables de la figure 3c et permettant d'évaluer les performances du procédé et du dispositif de traitement optimisé d'un signal perturbateur, objet de la présente invention.

They will be better understood on reading the description and on observing the drawings below in which, in addition to FIGS. 1a to 1d relating to the prior art,

• FIG. 2a represents, by way of nonlimiting example, an illustrative block diagram of the implementation of the method, object of the present invention, in the time domain;
• FIG. 2b represents, by way of nonlimiting example, an illustrative block diagram of the implementation of the method, object of the present invention, in the time domain, in the more specific case of the existence of a signal reception generating an echo signal making a specific contribution to the disturbing signal;
• FIG. 2c represents, by way of nonlimiting example, in a situation similar to that of FIG. 2a, a block diagram illustrating the implementation of the method, object of the present invention, in the frequency domain;
• FIG. 2d represents, by way of nonlimiting example, an illustrative block diagram of the implementation of the method, object of the present invention, in a situation similar to that of FIG. 2b, in the frequency domain, in the case particular of a reception signal generating an echo signal making a specific contribution to the disturbing signal;
• FIG. 2e represents, by way of nonlimiting example, an illustrative block diagram of a preferential implementation by a processing by successive blocks of the observation signal, in a situation similar to that of FIG. 2d, in the case the existence of a reception signal generating an echo signal making a specific contribution to the disturbing signal;
• FIG. 3a represents, in the form of block diagrams, the block diagram of a device allowing, in the frequency domain, the general processing, respectively the processing by successive blocks, of the observation signal, in the general case of existence a reception signal, generating an echo signal making a specific contribution to the disturbing signal;
• FIG. 3b represents an advantageous embodiment detail of a module for estimating the power spectral density of the useful signal more particularly implemented in the device represented in FIG. 3a, when, in particular, the block processing is implemented artwork ;
• FIG. 3c represents an alternative embodiment of the device represented in FIGS. 3a or 3b, in which a module for estimating the spectral density of the echo of a reception signal and a module for estimating the spectral density of the signal noise, in the context of an application to hands-free mobile radio, are introduced;
• Figures 3d and 3e show, by way of nonlimiting example, a module for estimating the power spectral density of the noise signal and the observation signal, by recursive filtering from a forgetting factor;
• FIGS. 4a to 4e represent different timing diagrams of signals, recorded at remarkable test points in FIG. 3c and making it possible to evaluate the performance of the method and of the device for optimized processing of a disturbing signal, object of the present invention.

The optimized signal processing method disruptive during a sound recording, in accordance with the object of the present invention will now be described in connection with Figures 2a to 2d.

Generally, it indicates that the signal above disturbance consists at least of a signal of noise, which, due to the very definition of a signal noise, is considered to be significantly decorrelated from the original useful signal that we wish to recover following attenuation, or even elimination, of this noise signal.

Firstly, it is indicated that the process of optimized processing of the disturbing signal, object of the present invention is made from a signal of observation, noted y (t), available in step 100 of origin in FIG. 2a, this observation signal being deemed to be formed from the original useful signal to be recovered, noted s (t) and the disturbing signal, denoted p (t).

More specifically, it indicates that the disturbing signal, in addition to the aforementioned noise signal, may have different contributions such as a signal echo, reverberation signal, or any other form of noise signal, as will be described later in the description. In the context of Figure 2a, we limit ourselves the existence of a substantially decorrelated noise signal from the useful signal, as mentioned previously.

In accordance with the process which is the subject of this invention, this consists in carrying out an estimation at step 101 of the disturbing signal to generate a signal estimated disruptor, noted p and (t). Of course, at the end of the aforementioned step 101, not only the signal is available estimated disturbance p and (t), but also of the observation signal y (t) previously mentioned.

Following the obtaining of the estimated disturbing signal p and (t) in step 101, the optimized treatment method, in accordance with the object of the present invention, consists in perform, in a step 102, from the observation signal abovementioned y (t), a rough estimate of the useful signal, the useful signal estimated, by convention, being deemed, in reason for the actual decorrelation of the original wanted signal and of the noise signal, consist of the difference between the signal of observation y (t) and the estimated disturbing signal p and (t). In end of step 102, an estimated useful signal is obtained, obtained following the rough estimation step, this useful signal estimated approximately corresponding to the useful signal of origin s (t) and for this reason noted s andu.

Following steps 101 and 102 above, the method of optimized processing, object of the present invention, then consists in performing a filtering 103 of the signal observation y (t) from the estimated disturbing signal p and (t) and optimal filtering to generate a signal useful, noted su.

As further shown in Figure 2a, the optimal filtering 103 then makes it possible to perform a minimization, in a step 104, of the error between the useful signal estimated su and the useful signal su. The whole process achieved by steps 103 and 104 through steps 101 and 102 then make it possible to obtain convergence, thanks to optimal filtering, the estimated useful signal s andu and useful signal su to the original useful signal s (t) for a substantially zero error between the useful signal su and the estimated useful signal s andu. The estimated useful signal s andu or the useful signal su is then substantially equal to the useful signal original s (t) to the nearest filtering errors.

FIG. 2a shows the method of optimized processing of a disturbing signal, in accordance with the object of the present invention, in the time domain. It is pointed out in particular that the notions of estimating the disturbing signal, rough estimation of the useful signal and optimal filtering can be perfectly defined in the time domain.

However, while in the case of Figure 2a, the observation signal y (t) is deemed to contain only one disturbing signal p (t) formed by a single noise signal significantly decorrelated from the useful signal, the process, object of the present invention can also, in a way particularly advantageous, be implemented when, at aforementioned observation signal corresponds to a disturbing signal p (t) to which is added, in addition to the noise signal substantially decorrelated from the original useful signal s (t), a signal echo, noted z (t). This echo signal corresponds, in particular in hands-free mobile phone situations for example, to a disturbing signal generated by a signal of observation, noted x (t), under conditions which will be explained in more detail later in the description.

Under these conditions, as shown in figure 2b, and always within the framework of an optimized treatment in the time domain, consistent with the subject of this invention, it is indicated that the estimation of the disturbing signal in step 101 advantageously consists in carrying out a separate estimate of the contribution 101b of this signal from reception and contribution 101a of the noise signal to this disturbing signal.

In Figure 2b, we find the same notations that in the case of FIG. 2a, the disturbing signal estimated being always noted p and (t) and consisting then, not only in the contribution of the decorrelated noise signal of the useful signal, in the same way as in the case of figure 2a, but also in the contribution to this signal disturbance of the reception signal noted x (t).

Due to the decorrelation between the signal reception and noise signal, particularly in one aspect advantageous of the process, object of the present invention, the process applied can then be substantially identical to that explained in connection with FIG. 2a.

For the same reason, it is indicated that the signal estimated disturbance p and (t) as well as the useful signal su play, in the optimal filtering process 103 and in the rough estimation process 102, respectively in the error calculation process and minimization of this error 104, the same role as in the case of FIG. 2a.

Under these conditions, and for the same reasons, the useful signal su resulting from optimal filtering in step 103 converges to the value of the estimated useful signal su and, in consequently, towards the value of the original useful signal s (t).

A preferred embodiment of the optimized processing of a disturbing signal in the field frequency corresponding to the case where the disturbing signal p (t) is simply constituted by a noise signal decorrelated from the useful signal s (t), respectively in the case where, on the contrary, this disturbing signal is constituted, not only by the contribution of a decorrelated noise signal of the useful signal, but also by the contribution of a reception signal x (t) such as an echo signal, a signal of reverberation or the like generated by the signal observation y (t), will be given in conjunction with the figures 2c, respectively 2d.

This preferred embodiment is particularly advantageous due in particular to the fact that in the part of an implementation by digital techniques of filtering in the frequency domain, it is not necessary to implement an echo canceller unlike to the techniques that have been described in relation to the prior art previously in the description.

In connection with FIG. 2c, and in the case where the disturbing signal p (t) is formed simply by a signal of decorrelated noise of the useful signal, the processing method optimized, object of the present invention, in the field frequency, may consist in performing in step 100 a frequency transform of the observation signal y (t) at using a Fourier transform, such as a transform fast, noted FFT in the usual way, to allow to generate a transformed signal Y (f), this signal being representative, in the frequency domain, of the signal observation.

In addition, the aforementioned step 100 consists in carrying out an estimation from the transformed signal Y (f) of a signal representative of the power spectral density of the observation signal, this signal being denoted γ and _yy (f).

At the end of step 100, we thus have not only the transformed signal Y (f) representative of the frequency transform of the observation signal y (t), but also the signal representative of the estimated power spectral density. of this observation signal, signal noted γ and _γγ (f).

According to a particularly advantageous aspect of the implementation of the optimized processing method of a disturbing signal, object of the present invention, it is indicated that the step 102 of estimation of the useful signal can then be carried out directly on the spectral density of estimated power, on the one hand, of the observation signal γ and _yy (f) and, on the other hand, of the signal representative of the estimated power spectral density of the disturbing signal obtained at the end of step 101, noted γ and _pp (f). In such a case, and in accordance with a remarkable aspect of the method according to the invention, the step 102 of rough estimation of the useful signal then amounts to carrying out an a posteriori estimation of the power spectral density of the useful signal, which, for this reason is noted γ and _ss (f). At the end of step 102, the signal representative of the estimated power spectral density of the aforementioned useful signal is then available.

According to another particularly advantageous aspect of the method, object of the present invention, when the processing is carried out in the frequency domain, as shown in FIG. 2c, the optimal filtering step 103 is carried out on the signal representative of the frequency transform of the observation signal Y (f) from signals representative of the estimated power spectral density of the disturbing signal γ and _pp (f) and of the signal representative of the estimated power spectral density of the useful signal, denoted γ and _ss (f ), available at the end of step 102 above. In this case, the optimal filtering step 103 and the step of calculating an error and minimizing this error 104 can be carried out by means of the same global filtering step, noted for this reason 103 + 104 on the FIG. 2c, the processing in the frequency domain, in particular the digital processing allowing, thanks to the implementation of a single optimal filter, the optimization of the useful signal, the error signal between the useful signal and the useful signal estimated, or more exactly between the estimated power spectral densities of these signals, being directly available due to the optimal filtering performed. For this reason, the global filtering is represented in dotted lines as the meeting of steps 103 and 104 in FIG. 2c.

Of course, if the disturbing signal p (t) consists, not only in the contribution of a signal noise, as described in relation to Figure 2c, but also in the contribution of a reception signal, and, analogously to the corresponding processing mode shown in Figure 2b, the process, subject of this invention, for treatment in the frequency domain, can of course be implemented with the same advantages that in the case of Figure 2c in the case of the presence of a reception signal, as shown in figure 2d.

In this case, the process which is the subject of the present invention, consists in performing a frequency transform of the observation signal, in step 100a, transformed denoted FFT, to generate the representative transformed signal in the frequency domain of the observation signal Y (f) as well as a frequency transform of the reception signal, in step 100b, to generate a transformed signal representative of the reception signal and denoted X (f).

Analogously to the process described in FIG. 2c, an estimation step is carried out in steps 100a and 100b, this estimation step consisting in obtaining, from each transformed signal Y (f) and X (f) previously cited, to obtain a signal representative of the estimated power spectral density of the observation signal, noted for this reason γ and _yy (f), respectively of the reception signal, noted for this reason γ and _xx (f).

In general, the density estimate power spectral of the observation signal, of the signal reception, the echo signal can be implemented at recursive filtering based on a forgetting factor, as will be described later in the description.

The estimation of the power spectral density of the disturbing signal carried out in step 101 consists in performing the step of estimating the power spectral density of the disturbing signal γ and _pp (f) on the signal representative of the spectral density. power of the observation signal γ and _yy (f) available at the end of step 100a, respectively on the signal representative of the power spectral density of the reception signal γ and _xx (f) available at the end of step 100b. We thus obtain, at the end of steps 101a and 101b, that is to say finally at the end of step 101, signals representative of the estimated power spectral density of the noise signal, signal noted γ and _ppy (f ), respectively of the echo signal generated by the reception signal noted for this reason γ and _ppx (f).

Due to the same principle of no correlation between the contribution of noise to the disturbing signal and the useful signal and the contribution of noise to the disturbing signal and the contribution of the reception signal to this same disturbing signal and this same useful signal, the density estimated power spectral resulting from the disturbing signal, denoted by this fact γ and _pp (f), is deemed to consist of the sum of the estimated power spectral densities γ and _ppy (f) and γ and _ppx (f).

Due to the unique notation used for the description of FIGS. 2d and 2c, step 102 as represented in FIG. 2d also consists in carrying out an estimation of the spectral density of the useful signal γ and _ss (f) then deemed to be equal. unlike the estimated spectral densities of the observation signal γ and _yy (f) and the disturbing signal γ and _pp (f).

Of course, and as in the case of FIG. 2c, the signals of estimated spectral density of the useful signal γ and _ss (f) available in step 102 and of the disturbing signal γ and _pp (f) then make it possible to ensuring the optimal filtering in step 103 and, more generally, the global filtering 103 + 104 on the signal Y (f) representative in the frequency domain of the observation signal.

With regard to the criterion for minimizing the error between the useful signal and the estimated useful signal, it is indicated that the minimization criterion can consist in minimizing the mean square error of estimation according to the relation (1): E [(su- s u) 2 ]

The above relation (1) can be used, either for processing in the time domain, i.e. for frequency domain treatment.

A justification of the entire process optimized processing, object of the present invention, will now given at the theoretical level for treatment in the frequency domain.

The minimization of the aforementioned error between the useful signal and the estimated useful signal leads, for the frequency domain, to the implementation of a filtering of the observation signal in its form of signal representative of the observation signal in the frequency domain Y (f), according to relation (2): S (f) = T (f) Y (f) = su.

γ and_ys(f)

désigne l'interspectre entre le signal d'observation, c'est-à-dire le signal représentatif du signal d'observation dans le domaine fréquentiel et le signal utile, et

γ and_yy(f)

désigne la densité spectrale de puissance estimée, ci-après désignée par dsp, du signal d'observation.

In this relation, T (f) represents the frequency response of an optimal filtering whose expression is given by equation (3): T (f) = γ ys (F) γ yy (F) . In this relationship,

γ and _ys (f)

designates the interspectrum between the observation signal, that is to say the signal representative of the observation signal in the frequency domain and the useful signal, and

γ and _yy (f)

denotes the estimated power spectral density, hereinafter referred to as dsp, of the observation signal.

γ and_ss(f)

désigne la densité spectrale de puissance estimée du signal utile,

γ and_pp(f)

désigne la densité spectrale de puissance estimée du signal perturbateur.

Taking into account the previously mentioned realistic hypotheses of effective decorrelation between the useful signal and the disturbing signal consisting of noise and echo, the frequency response of the optimal filtering verifies the relation (4): T (f) = γ ss (F) γ ss (f) + γ pp (F) In this relationship:

γ and _{ss (f)}

denotes the estimated power spectral density of the useful signal,

γ and _pp (f)

denotes the estimated power spectral density of the disturbing signal.

From a practical point of view, the estimated power spectral density of the useful signal γ and _{ss (f)} is not known a priori. This signal can for example be estimated taking into account the preceding hypotheses of decorrelation between the useful signal and the disturbing signal by using the process of spectral subtraction previously quoted, verifying the relation (5): γ ss (f) = γ yy (f) - γ pp (F).

The optimized signal processing process disruptive, in accordance with the object of the present invention, thus reduces to the implementation of a single filtering optimal, which allows to reduce overall all the components constituting the disturbing signal. Indeed, we understand in particular that the signal disruptive may consist of a plurality of components provided there is sufficient decorrelation between the useful signal and the disturbing signal, i.e. each components constituting the latter. This assumption is widely verified in various applications linked by example of hands-free telephony in vehicles cars, or hands-free video conferencing, and, more generally, to all types of applications in which a plurality of components of a signal disruptive can be highlighted.

Dans cette relation, P représente le nombre de composantes du signal perturbateur.In such a case, for a disturbing signal consisting of a plurality of components of this disturbing signal, the estimated power spectral density of the disturbing signal γ and _pp (f) is then taken equal to the sum of the spectral densities of estimated power γ and ⁱ _pp (f) of each component of rank i of this disturbing signal. In this case, the signal representative of the estimated power spectral density of the disturbing signal checks the relation (6):

In this relation, P represents the number of components of the disturbing signal.

A preferred embodiment of the optimized processing, object of the present invention, will now described in connection with FIG. 2e in the case where block processing of the observation signal is realized.

In the context of such treatment, it is understood that particular that the observation signal y (t) available is of course sampled at a sampling frequency adequate, successive samples being subdivided in sample blocks. At each block of samples is assigned a successive rank m, m denoting in fact the rank of the current block subjected to the treatment. We understand in particular that the technique of building up sample blocks is a classic technique, successive blocks samples that may be subject to some recovery typically equal to 50% in number of constituent samples of each block.

In the context of Figure 2e, treatment with blocks is deemed to be carried out in the most general manner when the disturbing signal takes into account, not only the contribution of a noise signal, but also that generated by a reception signal x (t).

As shown in FIG. 2e, in step 100a, in addition to the subdivision of the observation signal into successive blocks of rank m, each block of samples being denoted Bm (t) is of course subjected to a frequency transformation FFT allowing d '' obtain blocks of samples in the frequency domain denoted Bm (f). Step 100a also consists in carrying out an estimation of the power spectral density of the observation signal on the current block, the signal of the estimated power spectral density of the observation signal being noted γ and _yy (f, m) where m naturally denotes the index relating to the current block.

At the end of step 100a, in fact, not only is the signal representative of the estimated power spectral density of the aforementioned observation signal γ and _yy (f, m), but also the representative block Bm (f) of the observation signal for the current block of rank m considered.

It is the same in step 100b for which, by analogy with FIG. 2d, a corresponding processing is applied to the reception signal x (t), this processing then consisting of a subdivision into corresponding blocks of rank m, each block being denoted B'm (t), each aforementioned block being subjected to a frequency transformation, denoted FFT, this operation making it possible to obtain blocks representative of the blocks of samples in the frequency space and denoted for this reason B'm (f). Step 100b shown in FIG. 2e also includes an operation of estimating the power spectral density of the reception signal on the current block B'm (f). At the end of step 100b in FIG. 2e, there is each current block B'm (f) representative of the sample block in the frequency domain and a signal representative of the estimated power spectral density of the reception signal for the aforementioned current block, this signal being noted γ and _xx (f, m).

As also shown in FIG. 2e, the optimized processing method, in accordance with the object of the present invention, then consists, in step 101, in carrying out an estimation of the power spectral density of each component of the signal disruptor previously cited γ and ⁱ _pp (f, m). It is understood, for example, that the signal representative of the power spectral density of each component of the disturbing signal γ and ⁱ _pp (f, m) is in fact constituted at least by the signal representative of the estimated power spectral density γ and _ppy ( f, m) representative of the contribution of the noise signal to the disturbing signal and by the signal representative of the estimated power spectral density of the contribution of the reception signal to this disturbing signal γ and _ppx (f, m).

The estimation of the power spectral density of each component of the disturbing signal γ and ⁱ _pp (f, m) is thus carried out from the reception signal and, more particularly, from the estimated power spectral density of the reception signal γ and _xx (f, m) and of the current block B'm (f), of the estimation of the power spectral density of the observation signal on the current block Bm (f) of the observation signal of the same rank m.

At the end of step 101, in FIG. 2e, there is in fact, for the current block of rank m of the observation signal and of the reception signal, the estimated power spectral density of the observation signal on this block current noted γ and _yy (f, m) and, of course, an estimate of the power spectral density of the disturbing signal γ and _pp (f, m), which of course checks the previous relation (6).

As shown in FIG. 2e, the power spectral density of the useful signal is then estimated on the current block by a so-called posterior estimation. The signal representative of the estimated power spectral density of the useful signal then checks the relation (7):

Remember that the notion of a posteriori estimate covers the notion of spectral density estimation of useful signal strength in the absence of any knowledge of the last. This operation bears the reference 102a on Figure 2e.

The posterior estimation operation 102a is then followed by a step 102b of a priori estimation of the amplitude of the spectrum of the useful signal on the current block. In general, we indicate that the amplitude of the spectrum of the useful signal on the current block satisfies the general relation (8): AT ss (f, m) = T (f, m). Y (f, m).

T(f,m)

désigne la réponse en fréquence du filtrage optimal pour le bloc courant ;

Y(f,m)

désigne la transformée de fréquentielle à court terme, c'est-à-dire la transformée de Fourier, sur le bloc courant du signal d'observation.

In this relationship:

T (f, m)

designates the frequency response of the optimal filtering for the current block;

Y (f, m)

denotes the short-term frequency transform, that is to say the Fourier transform, on the current block of the observation signal.

It is indicated in particular that the signal Y (f, m) can be obtained from the current block Bm (t) and application of a simple short-term Fourier transform on this block current to obtain the signal Y (f, m).

In order to carry out an a priori estimate of the amplitude of the spectrum of the useful signal, it is indicated that this operation, carried out in step 102b, consists in taking as value the signal corresponding to the filtering of the current block of the observation signal by the memorization of the value of the frequency response of the optimal filtering calculated on the preceding block, that is to say T (f, m-1), according to the relation (9): AT ss (f, m) = T (f, m-1) .Y (f, m). It is thus understood that the estimation step 102b can be summed up by storing the value of the frequency response of the optimal filtering calculated on the previous block.

The aforementioned step 102b is then followed by the estimation of the power spectral density of the useful signal in step 102c shown in FIG. 2e. In the aforementioned step 102c, the estimated power spectral density of the useful signal is established so as to verify the relation (10) below: γ ss (f, m) = β (m) | A ss (F, m) | 2 + (1-β (m)) γ ss-post (F, m).

Step 102c of estimating the power spectral density of the useful signal is carried out thanks to the implementation of a step 102d making it possible to generate, for each current block Bm (f), a weighting parameter β (m ) allowing a suitable weight to be assigned between the current estimate made from the filtering applied to the previous block of rank m-1 and the contribution for the current frame of the estimated power spectral density of the useful signal, which is of course represented by the signal γ and _ss-post (f, m).

At the end of step 102, there is of course the signal representative of the estimated power spectral density of the useful signal, noted γ and _ss (f, m). The optimal filtering process can then be controlled for the current block on the signal Y (f, m) thanks to the global filtering previously described in relation to FIG. 2d in steps 103 and 104. Of course, the transition to the next block is carried out by the increment m = m + 1 shown in Figure 2e.

A more detailed description of an embodiment non-limiting of an optimized treatment device a disturbing signal when taking sound from an observation signal, this signal being formed by a signal useful and this disturbing signal will now be described in conjunction with Figures 3a and 3b.

More specifically and because of the benefits major previously mentioned in the description in concerning the frequency treatment, the device, object of the present invention, represented in FIG. 3a, will be described for such treatment.

In addition, the disturbing signal is considered to be consisting of a noise and an echo generated by a signal reception. In the same way as in the case of FIGS. 2c and 2d, the observation signal is noted y (t) and is considered supplied by an M microphone, and the reception signal denoted x (t) corresponds to that of the signal delivered to a loudspeaker HP in the context of mobile radiotelephony hands free for example. We understand that in the passenger compartment, the loudspeaker and the microphone M being necessarily close to each other, the contribution to the signal disturbing the reception signal cannot in any case be neglected, while of course other components such as engine noise from the vehicle, traffic noise generated by traffic neighbor for example constitute as many components and contributions to the disturbing signal.

Description of Figure 3a and Figure 3b is given in the case of the general principle of treatment overall as well as in the case of a similar treatment performed in the form of block processing, the references constituent elements of the treatment device optimized, object of the present invention, in the case of block processing, corresponding to those allocated for general treatment, however with an index m corresponding to the row designation of the current block considered, as previously described in connection with the Figure 2d and 2e.

As shown in FIG. 3a, the observation signal y (t) delivered by the microphone M is subjected by means of a module, denoted T ₁ (f, m), T ₁ (f) , to digital sampling at an appropriate frequency, to a subdivision by blocks and of course to a frequency transform, denoted FFT in FIG. 3a. The module T ₁ (f, m) then delivers the signal Y (f, m) representative in the frequency domain of the observation signal on the block of rank m considered.

The same applies to the reception signal via a module T ₂ (f, m), T ₂ (f), which makes it possible to deliver the representative signal in the frequency domain X (f, m) and the blocks B'm (f) representative of the reception signal for the block of rank m considered.

The modules T ₁ (f, m) and T ₂ (f, m) are modules of the conventional type, identical, synchronized by the same clock signal, not shown. As such, these modules will not be described in detail since they correspond to modules normally used in the corresponding technical field and, as such, perfectly known to those skilled in the art.

As will be further observed in FIG. 3a, the optimized processing device, object of the present invention, comprises a module 1.1 _m for estimating the power spectral density of the observation signal delivering, at from this observation signal, or, more precisely, from the signal representative in the frequency domain of this observation signal, that is to say either the signal Y (f), or the signal Y (f, m), a digital signal representative of the estimated power spectral density of the observation signal and for this reason noted, for the same reason, γ and _yy (f), respectively γ and _yy (f, m) on the current block m considered.

In addition, the device according to the invention, as shown in FIG. 3a, comprises a module 2.2 _m for estimating the power spectral density of the disturbing signal receiving the reception signal, or, more precisely, the signal representative in the frequency domain of this reception signal, that is to say either the signal X (f, m), or the signal X (f). The module 2 for estimating the power spectral density of the disturbing signal also receives the digital signal representative of the estimated power spectral density of the observation signal, that is to say the signal γ and _yy (f), respectively γ and _yy (f, m). It therefore delivers a digital signal representative of the estimated power spectral density of the disturbing signal, designated by γ and _pp (f). In a particular nonlimiting embodiment, it is indicated that the module 2.2 _m in fact delivers all the signals representative of the estimated spectral power density of the components of the disturbing signal designated by γ and ⁱ _pp (f), respectively γ and ⁱ _pp (f, m).

A 3.3 _m module for estimating the power spectral density of the useful signal is also provided, which receives the digital signal representative of the estimated power spectral density of the observation signal γ and _yy (f), respectively γ and _yy (f, m) delivered by the module 1.1 _{m as} well as the digital signal representative of the estimated power spectral density of the disturbing signal γ and _pp (f), respectively γ and _pp (f, m) or the components of the latter, as previously mentioned. The 3.3 _m module for estimating the spectral power density of the useful signal delivers by a process inspired by the general principle of spectral subtraction a digital signal, noted γ and _ss (f), respectively γ and _ss (f, m ) representative of the estimated power spectral density of the aforementioned useful signal.

Finally, the device for optimized processing of a disturbing signal, object of the present invention, as shown in FIG. 3a, comprises a global filtering module, noted 4.4 _m , making it possible to ensure optimal filtering of the representative signal in the frequency domain of the observation signal, that is to say the signal Y (f) respectively Y (f, m) delivered by the module T ₁ (f, m), T ₁ (f).

As shown more specifically in FIG. 3a, the 4.4 _m filter module advantageously comprises a calculation module, denoted 4a, 4a _m , of the coefficients of an optimal filter receiving the digital signal representative of the spectral density of estimated power of the disturbing signal γ and _pp (f), respectively γ and _pp (f, m), as well as the digital signal representative of the estimated power spectral density of the useful signal γ and _ss (f), respectively γ and _ss ( f, m). The module 4a, 4a _m shown in FIG. 3a delivers a digital filtering adaptation signal, denoted af , representative of an optimal filtering frequency response, verifying the relation (4) previously given in the description. It will of course be understood that in this relation, the estimated power spectral density of the disturbing signal corresponds to the sum of the spectral densities of the components of the disturbing signal according to relation (6) previously given in the description.

Finally, constituting the global filter module 4.4 _m , a module 4b, 4b _m receives the signal representative of the frequency response, that is to say the signal af delivered by the module 4a, 4a _m , to deliver , from the representative signal in the frequency domain of the observation signal, the useful signal su. It is understood in particular that the optimal filter module 4b, 4b _m can consist, for example, of a Wiener filter module. The signal delivered by this filter module 4b, 4b _m is then received by a reverse frequency transform module, for this reason noted FFT ^-1 , and by block synthesis, bearing the reference 5.5 _m , which delivers, from of the optimal filtering signal, the useful signal proper su (t) reconstituted in the time domain.

A more detailed description of a preferred embodiment of the 3 _m module shown in FIG. 3a for estimating the power spectral density of the useful signal corresponding to the mode of implementation of the method, object of the present invention, as shown in FIG. 2e, will now be given in conjunction with FIG. 3b for processing by blocks of successive rank m.

Of course, and in accordance with the description given in connection with FIG. 3a, the device which is the subject of the present invention comprises, in addition to the module T ₁ (f, m) delivering a succession of successive current blocks of rank m, the module of estimation of the power spectral density of the observation signal on the current block γ and _yy (f, m), module 1 _m , and the module of estimation of the power spectral density of each component of the disturbing signal γ and ¹ _pp (f, m), module 2 _m , the module for block estimation of the spectral power density of the useful signal, module 3 _m , which advantageously comprises, as shown in FIG. 3b, a module 30 _m for estimation a posteriori of the power spectral density of the useful signal on the current block, noted γ and _ss-post (f, m) verifying the relation (7) previously mentioned in the description. In addition, the 3 _m module also includes a 31 _m module for a priori estimation of the amplitude of the spectrum of the useful signal on the current block, verifying the relation (9) previously mentioned in the description. The 31 _m module receives, on the one hand, the signal γ and _ss-post (f, m) delivered by the module 30 _{m as} well, on the other hand, the signal Y (f, m) delivered by the block T ₁ (f, m), as well as a signal representative of the frequency response of the optimal filtering for the block preceding the current block, ie T (f, m-1) delivered for example by the block 4a _m of FIG. 3a .

The block 31 _m then delivers an a priori estimate of the amplitude of the spectrum of the useful signal denoted A _ss (f, m).

Finally, a module for calculating the power spectral density of the useful signal, for the current block, module 32 _m , is provided, which receives the a priori estimation signal of the amplitude of the spectrum of the useful signal spectrum A _ss (f , m) delivered by the module 31 _m as well as a signal representative of a weighting coefficient or parameter β (m) from a module 33 _m shown in FIG. 3b. The parameter β (m) makes it possible to assign a suitable weight between the estimate made at the previous block of rank m-1 and the contribution for the current frame of the power spectral density of the useful signal, as mentioned previously in the description . The parameter β (m) can be adjusted according to the characteristics of the useful signals and of the estimated noise. The module 32 _m then delivers the signal representative of the estimated power spectral density of the useful signal, verifying the relation (10) previously mentioned in the description.

The embodiment of the processing device optimized of a disturbing signal, object of this invention, as shown in Figures 3a and 3b, is not limiting.

It is understood in particular that in connection with the context of FIG. 2d for example, for a disturbing signal formed by an echo signal of this reception signal and of a noise signal, when the noise signal is substantially decorrelated of the echo signal and the module for estimating the power spectral density of the 2.2 _m echo signal then delivers a digital signal representative of the estimated power spectral density of the echo signal, denoted γ and _zz (f), respectively γ and _zz (f, m), the device, object of the present invention, is modified according to FIG. 3c where however the same references represent the same elements as in the case of FIG. 3a.

In such a hypothesis and taking into account the realistic hypothesis of decorrelation between the components of the disturbing signal, that is to say between the noise signal and the acoustic echo, the relation (4) previously mentioned in the description becomes the relation (11): T (f) = γ ss (f) γ ss (f) + γ bb (f) + γ zz (F) This relation represents the frequency response of the global filter taking into account the estimation of the power spectral density of the useful signal, the noise signal and the echo signal, denoted γ and _ss (f), respectively γ and _bb ( f, m), γ and _zz (f, m), with reference to Figure 3c.

In the same way, and due to the same realistic hypotheses of decorrelation between the components of the disturbing signal, the relation (5) previously mentioned in the description is transformed into the relation (12): γ ss (f) = γ yy (f) - γ bb (f) - γ zz (F).

In an advantageous embodiment of the device optimized processing of a disturbing signal, subject of the present invention, and in the more specific context of hands-free mobile telephony, an estimate of the power spectral density of single noise can be obtained in particular in the absence of an echo signal and useful signal.

In the same way, it is possible to estimate the power spectral density of the echo signal from the representative signal in the frequency domain of the signal reception and observation signal. For exemple nonlimiting, this estimate may involve a estimation of the transfer function of the acoustic channel between the reception signal and the observation signal.

Taking into account the preceding remarks, as shown in FIG. 3c, the device in such a case comprises, associated with the module 1.1 _m for estimating the power spectral density of the observation signal, an additional module for estimating the power spectral density of the noise affecting this observation signal.

In this case, moreover, as shown in FIG. 3c, the module 2.2 _m for estimating the power spectral density of the disturbing signal constitutes in fact a module for estimating the power spectral density of the acoustic echo, which delivers a signal representative of the estimated power spectral density of the acoustic echo, noted γ and _zz (f, m).

Under these conditions, as shown in FIG. 3c, the module for calculating the coefficients of the optimal filter 4a, 4a _m , directly receives the signal representative of the estimated power spectral density of the acoustic echo γ and _zz (f, m) , the signal representative of the estimated power spectral density of the noise, denoted γ and _bb (f, m) and, of course, the signal representative of the estimated power spectral density of the observation signal, denoted γ and _yy (f , m).

• du signal représentatif de la densité spectrale de puissance estimée γ and_yy(f), respectivement γ and_yy(f,m), délivré par le module 1,1_m,
• du signal représentatif de la densité spectrale de puissance estimée du bruit γ and_bb(f), respectivement γ and_bb(f,m),
• du signal représentatif de la densité spectrale de puissance γ and_zz(f), respectivement γ and_zz(f,m) délivré par le module 2,2_m,

le module 3,3_m d'estimation de la densité spectrale de puissance du signal utile γ and_ss(f), respectivement γ and_ss(f,m) n'est plus indispensable, le signal représentatif de la densité spectrale de puissance estimée du signal utile étant alors donné directement par la relation (12). La réponse en fréquence du filtre optimal, module 4b,4b_m est alors donnée par la relation (11) par l'intermédiaire du signal af précédemment mentionné dans la description.Under these conditions, and taking into account the availability at the level of module 4a, 4a _m of the aforementioned signals, that is to say:

• the signal representative of the estimated power spectral density γ and _yy (f), respectively γ and _yy (f, m), delivered by the module 1.1 _m ,
• of the signal representative of the estimated power spectral density of the noise γ and _bb (f), respectively γ and _bb (f, m),
• the signal representative of the power spectral density γ and _zz (f), respectively γ and _zz (f, m) delivered by the module 2.2 _m ,

the 3.3 _m module for estimating the power spectral density of the useful signal γ and _ss (f), respectively γ and _ss (f, m) is no longer essential, the signal representative of the estimated power spectral density of the useful signal then being given directly by the relation (12). The frequency response of the optimal filter, module 4b, 4b _m is then given by equation (11) via the signal af previously mentioned in the description.

In a specific embodiment of the device for optimizing the processing of a disturbing signal, object of the present invention, as shown in FIG. 3c, it is indicated that the estimation module 1a, 1a _m of the spectral density of the noise signal may advantageously include, as shown in FIG. 3d, a module for detecting the absence of useful signal and the absence of echo signal in the observation signal, and a first-order recursive filter having a factor of forgetting λ _bb , this forgetting factor consisting of a real coefficient between the value 0 and 1. In such a case, the recursive filter delivers the digital signal representative of the estimated power spectral density of the noise signal γ and _bb (f), respectively γ and _bb (f, m) verifying the relation (13): γ bb (f, m) = λ bb . γ bb (f, m-1) + (1-λ bb ) (| b (f, m) | 2 ).

In the above-mentioned relation (13), it is indicated that b (f, m) designates the frequency transform, Fourier transform, of the observation signal established over a current time segment of the observation signal in the absence of vocal activity , that is to say speech of one or the other of the two speakers in communication. As will be observed in FIG. 3d, the estimation module 1 _am , in its version relating to block processing, described in a nonlimiting manner, includes the voice activity detection module 10 _am receiving for example the signal Y (f, m) delivered by the module T ₁ (f, m), a switch controlled 11 _am by the voice activity detector module 10 _am , a square elevation module 12 _am , a multiplier circuit 13 _am receiving the signal delivered by the square elevation module 12 _am and the value 1-λ _bb . A summer 14 _am receives the signal delivered by the module 12 _am , delivers the signal representative of the estimated power spectral density of the noise signal γ and _bb (f, m) and receives by a feedback loop the signal representative of the density estimated power spectral of the noise signal γ and _bb (f, m-1) relating to the block preceding the current block via a 15 _am delay module, memory for example, and a weighting multiplier module 16 _am receiving the value λ _bb . On detection of absence of voice activity, the block B _m (f) delivered by the module T ₁ (f, m) corresponds to the frequency transform b (f, m) of the noise signal.

Finally, with regard to the module for estimating the power spectral density of the observation signal, in particular the 1.1 _m module, it is indicated that this can include, as shown in FIG. 3e, a filter first-order recursive, having a forgetting factor λ _yy consisting of a real coefficient between 0 and 1. The above-mentioned recursive filter then delivers the digital signal representative of the estimated power spectral density of the observation signal γ and _yy ( f), respectively γ and _yy (f, m), verifying the relation (14): γ yy (f) = λ yy . γ yy (f) + (1-λ yy ). | Y (f) | 2 . In this relation, Y (f), respectively Y (f, m), designates the signal representative in the frequency domain of the observation signal, i.e. the frequency transform of this observation signal on the block current for example.

The recursive filter represented in FIG. 3e comprises elements similar to those represented in FIG. 3d, the notations am being modified in m respectively, the value λ _yy being adapted accordingly.

Figures 4a to 4e make it possible to evaluate the performances obtained thanks to the implementation of the process for processing an optimized disturbing signal and using of a device, in accordance with the object of the present invention, as shown for example in Figure 3c.

In Figures 4a, 4b and 4c, the x-axis is graduated in seconds and the ordinate axis in value amplitude in PCM digital coding, 16-bit coding corresponding to a maximum value of 32,768.

The application context related to radiotelephony hands-free in a motor vehicle.

The signal sampling frequency was at a value of 8 kHz, the digital coding of the samples thus obtained being based on the PCM format, ie 16 bits linear.

During these tests, the signal broadcast on the speaker, receive signal, and microphone signal, i.e. the observation signal, have been recorded synchronously, the vehicle engine being stopped.

As part of this evaluation, noise and local speech recorded separately in a same vehicle were artificially summoned to the signal echo.

The original echo signal, picked up by the microphone M, is shown in Figure 4a.

The noisy observation signal, obtained as well as previously mentioned, is shown in Figure 4b, when the local voice, that is to say the speaker of the vehicle, was artificially disturbed by a noise signal and an echo signal corresponding to a human voice.

In FIGS. 4a and 4b, the signal represented by slots under the above records represents the voice activity detection in reception, i.e. on the reception signal received by the speaker HP.

The test observation signal represented in figure 4b thus comprises periods of noise alone, periods echo alone in the noise, but also periods of double talk, periods during which the two speakers in correspondence speak at the same time. The signal from test corresponds to a typical case in a mobile radio context hands free.

The characteristics of the observation signal are given in the table below: Average echo signal ratio (dB) 9.00 Signal to maximum echo ratio (dB) 38.61 Signal to minimum echo ratio (dB) -23.66 Standard deviation of signal to echo ratio (dB) 5.31 Average signal-to-noise ratio (dB) 6.17 Maximum signal-to-noise ratio (dB) 19.18 Minimum signal-to-noise ratio (dB) -27.38 Signal-to-noise ratio standard deviation (dB) 5.21

• longueur de la fenêtre d'analyse : 256 échantillons ;
• type de fenêtre d'analyse : fenêtre de Hanning ;
• recouvrement : 50%, soit 128 échantillons ;
• nombre de points de la transformée de Fourier rapide FFT : 256 points ;
• contrainte de convolution linéaire pour le filtrage réalisée par FFT inverse sur 512 points ;
• méthode de synthèse du signal : OLA, pour désigner la méthode Overlapp Add.

During these tests, in addition to the aforementioned sampling frequency, the processing parameters were as follows:

• length of the analysis window: 256 samples;
• type of analysis window: Hanning window;
• recovery: 50%, or 128 samples;
• number of points of the fast Fourier transform FFT: 256 points;
• linear convolution constraint for filtering performed by inverse FFT on 512 points;
• signal synthesis method: OLA, to denote the Overlapp Add method .

FIG. 4c represents the useful signal obtained in output of the device, the signal su of FIG. 3c. We finds an effective reduction in signal influence disruptive sensed when taking sound. Noise and original echo signal are greatly attenuated by the application of treatment.

In order to assess the reduction brought by the treatment on noise and on echo, we have represented in figures 4d and 4e, on the one hand, the attenuation of the echo in decibels, and, on the other hand, the attenuation of noise in decibels.

The echo attenuation is evaluated by an energy measurement, known as ERLE, for Echo Return Loss Enhancement, this measurement being evaluated on blocks of 256 samples in the absence of recovery.

Likewise, noise attenuation is assessed on blocks of 256 samples without overlap.

The analysis of Figures 4d and 4e shows that the process and the optimized treatment device, object of the present invention, reduce power average of the acoustic echo picked up by the microphone M, of around 15 dB during the single echo periods and around 10 dB during periods of double talk.

Regarding power reduction average noise, this reduction is around 18 dB during the noise period alone. During echo periods single and double talk, the optimized overall processing automatically adapts to the observation signal delivered by the microphone M. Indeed, we can then observe a 15 dB reduction in noise power during periods echo alone and 8 dB during periods of double talk.

The optimized treatment method and device disturbing signals, objects of the present invention, appear very advantageous insofar as they allow to reduce the distortions introduced on the signal useful local speaking. In addition, reducing the attenuation to the echo signal and noise signal during periods of vocal activity in transmission does not introduce undesirable effects on the signal transmitted to the correspondent distant because the echo signal and the noise signal remaining at the end of treatment are then subjectively masked by the local speech signal.

The method and the device, objects of the present invention, are particularly well suited to radiotelephony hands-free mobile in motor vehicles. Indeed, while some European countries have already taken measures to ban the use of a handset classic cell phone while driving a motor vehicle, we should expect a generalization such measures. Analysis of hands-free telephony in vehicles helped to highlight the two main factors discomfort to the driver, corresponding not only to the simultaneous communication, but also ambient noise level, whereas for the correspondent of the latter, the most important genes are caused by the presence of noise and an acoustic echo, induced by the acoustic coupling existing between transducers.

By implementing a comprehensive treatment of disturbing signal, the method and the device, objects of the invention, while ensuring a sufficient quality of speech, allow to get rid of the implementation of a adaptive acoustic echo cancellation system, of which implantation is particularly expensive and difficult to settle.

Claims (12)

1. Method of optimized processing of a disturbing signal consisting at least of a noise signal during a sound capture, on the basis of an observation signal formed of a starting useful signal and of this disturbing signal, characterized in that it consists in performing:

   an estimation of the disturbing signal so as to generate an estimated disturbing signal;

   an estimation of the useful signal so as to generate an estimated useful signal, by subtracting the estimated disturbing signal to the observation signal;

   a filtering of the said observation signal on the basis of the said estimated disturbing signal and of an optimal filtering so as to generate a useful signal, the said optimal filtering making it possible to minimize the error between the said useful signal and the said estimated useful signal, the said estimated useful signal converging towards the said starting useful signal for a substantially zero error between the said useful signal and the said estimated useful signal.
2. Method according to Claim 1, characterized in that, when the sound capture is performed in the presence of a reception signal, the said estimation of the disturbing signal consists in performing a separate estimation of the contribution of this reception signal and of the contribution of the noise signal of the disturbing signal.
3. Method according to one of Claims 1 or 2, characterized in that, for a processing of this disturbing signal in the frequency domain, the latter consists:

   in performing a frequency transform of the observation signal, respectively of the reception signal, so as to generate a transformed signal which is representative, in the frequency domain, of the observation signal, respectively of the reception signal;

   in estimating, on the basis of each transformed signal, a signal representative of the power spectral density of the observation signal, respectively of the reception signal;

   in applying the step for estimating the disturbing signal to the signal representative of the power spectral density of the observation signal, respectively to the signal representative of the power spectral density of the reception signal;

   in applying the said optimal filtering to the said transformed signal representative of the observation signal, so as to generate a transformed signal representative of the useful signal.
4. Method according to Claim 3, characterized in that the said optimal filtering is carried out on the basis of a signal representative of the estimated power spectral density of the useful signal, derived via a spectral subtraction procedure and satisfying the relation: γ ss(f)=γ yy(f)-γ pp(f) in which:

   γ and_yy(f) designates the estimated power spectral density of the observation signal;

   γ and_pp(f) designates the estimated power spectral density of the said disturbing signal.
5. Method according to Claim 3 or 4, characterized in that, for a disturbing signal consisting of a plurality of components of this disturbing signal, the estimated power spectral density of the disturbing signal γ and_pp(f) is taken equal to the sum of the estimated power spectral densities γ andⁱ _pp(f) of each component of rank i of this disturbing signal and satisfies the relation:

   

   where P represents the number of components of the disturbing signal.
6. Method according to Claims 4 or 5, characterized in that, for a block processing operation in the frequency domain of the said observation signal, this signal being subdivided into blocks of successive samples, the said method, for every current block of rank m, with a view to deriving the said estimated power spectral density of the useful signal, consists in performing:

   an estimation of the power spectral density of the observation signal over the current block γ and_yy(f,m);

   an estimation of the power spectral density of each component of the disturbing signal γ andⁱ _pp(f,m), on the basis of the said reception signal, of the current block of rank m of the observation signal and of the estimation of the power spectral density of the observation signal over the current block γ and_yy(f,m);

   an a-posteriori estimation of the power spectral density of the useful signal over the current block, γ and_ss-post(f,m) satisfying the relation:

   

   an a-priori estimation of the amplitude of the spectrum of the useful signal over the current block satisfying the relation:

   Ass (f,m)=T(f,m-1) . Y(f,m) where

   T(f,m-1)

   designates the frequency response of the said optimal filtering applied to the preceding block;

   Y(f,m)

   designates the short-term Fourier transform, over the current block, of the said observation signal,

   the said estimated power spectral density of the useful signal satisfying, for the current block, the relation: γ ss(f,m) = β(m)|Ass(f,m)|2 + (1-β(m))γ ss-post(f,m) in which relation β(m) designates, for the current block, a weighting parameter making it possible to assign a matched weight between the current estimation performed on the basis of the filtering applied to the preceding block, of rank m-1, and the contribution in respect of the current frame of the power spectral density of the useful signal.
7. Device for optimized processing of a disturbing signal during a sound capture, on the basis of an observation signal, formed of a useful signal and of this disturbing signal, the said disturbing signal consisting of a noise and an echo generated by a reception signal, characterized in that, for a processing operation in the frequency domain of these signals, the said device comprises at least:

   means for estimating the power spectral density of the said observation signal which deliver, on the basis of the said observation signal, a digital signal representative of the estimated power spectral density of the said observation signal γ and_yy(f);

   means for estimating the power spectral density of the said disturbing signal which receive the said reception signal and the said digital signal representative of the estimated power spectral density of the said observation signal γ and_yy(f) and deliver a digital signal representative of the estimated power spectral density of the said disturbing signal γ and_pp(f);

   means for estimating the power spectral density of the useful signal which receive the said digital signal representative of the estimated power spectral density of the said observation signal γ and_yy(f) and the said digital signal representative of the estimated power spectral density of the said disturbing signal γ and_pp(f) and deliver, via spectral subtraction, a digital signal representative of the estimated power spectral density of the useful signal γ and_ss(f);

   means for computing the coefficients of an optimal filter which receive the said digital signal representative of the estimated power spectral density of the said disturbing signal γ and_pp(f) and the said digital signal representative of the estimated power spectral density of the useful signal γ and_ss(f) and deliver a filtering adaptation digital signal representative of a filtering frequency response of the form: T(f)= γ ss(f) γ ss(f)+γ pp(f) ;

   means for optimal filtering which receive the said observation signal and the said filtering adaptation digital signal and deliver the said estimated useful signal representative of the said useful signal.
8. Device according to Claim 7, characterized in that, for a disturbing signal consisting of a plurality of components of the disturbing signal, the said means for estimating the power spectral density of the useful signal receive the said digital signal representative of the estimated power spectral density of the said observation signal γ and_yy(f) and the said digital signal representative of the estimated power spectral density γ andⁱ _pp(f) of the various components of the disturbing signal and deliver a digital signal representative of the estimated power spectral density of the useful signal γ and_ss(f).
9. Device according to Claim 8, characterized in that, for a block processing operation in the frequency domain of the said observation signal, the said device comprises:

   means for subdividing the said observation signal into successive blocks which receive this observation signal and deliver a succession of successive current blocks of rank m;

   means for estimating the power spectral density of the observation signal over the current block γ and_yy(f,m);

   means for estimating the power spectral density of each component of the disturbing signal γ andⁱ _pp(f,m), on the basis of the said reception signal, of the current block of rank m of the observation signal and of the estimation of the power spectral density of the observation signal over the current block γ and_yy(f,m);

   means of blockwise estimation of the power spectral density of the useful signal comprising:

   means of a-posteriori estimation of the power spectral density of the useful signal over the current b
   lock, γ and_ss-post(f,m) satisfying the relation:

   

   means of a-priori estimation of the amplitude of the spectrum of the useful signal over the current block satisfying the relation:

   Ass(f,m)=T(f,m-1).Y(f,m) where

   T(f,m-1)

   designates the frequency response of the said optimal filtering applied to the preceding block;

   Y(f,m)

   designates the short-term Fourier transform, over the current block, of the said observation signal,

   the said estimated power spectral density of the useful signal satisfying, for the current block, the relation: γ ss(f,m)=β(m)|Ass(f,m)|2+(1-β(m))γ ss-post(f,m) in which relation β(m) designates, for the current block, a weighting parameter making it possible to assign a matched weight between the current estimation performed on the basis of the filtering applied to the preceding block, of rank m-1, and the contribution in respect of the current frame of the power spectral density of the useful signal.
10. Device according to Claim 7, characterized in that, for a disturbing signal formed by an echo signal of this reception signal and of a noise signal, the said noise signal being substantially uncorrelated from the said echo signal and the said means for estimating the power spectral density of the echo signal delivering a digital signal representative of the estimated power spectral density of the echo signal γ and_zz(f), this device moreover comprises means for estimating the power spectral density of the noise signal which deliver to the said means for computing the coefficients of an optimal filter a digital signal representative of the estimated power spectral density of the noise signal γ and_bb(f), the said computing means delivering a filtering adaptation digital signal representative of a filtering frequency response of the form: T(f)= γ ss(f) γ ss(f)+γ bb(f)+γ zz(f) with γ ss(f)=γ yy(f)-γ bb(f)-γ zz(f).
11. Device according to Claims 7, 9 and 10, characterized in that the said means for estimating the power spectral density of the noise signal comprise:

    a means for detecting the absence of a useful signal and the absence of an echo signal in the observation signal;

    a first-order recursive filter having a neglect factor λ_bb, a real coefficient lying between 0 and 1, the said recursive filter delivering the said digital signal representative of the estimated power spectral density of the noise signal γ and_bb(f) of the form: γ bb(f,m)=λbb.γ bb(f,m-1)+(1-λbb)(|b(f,m)|2)

    where b(f,m) designates the Fourier transform of the observation signal, derived over a current time segment of the observation signal in the absence of voice activity.
12. Device according to one of Claims 7 to 11, characterized in that the said means for estimating the power spectral density of the observation signal comprise:

    a first-order recursive filter having a neglect factor λ_yy, a real coefficient lying between 0 and 1, the said recursive filter delivering the said digital signal representative of the estimated power spectral density of the observation signal γ and_yy(f) of the form:

    γ yy(f)=λyy.γ yy(f)+(1-λyy).|Y(f)|2 where Y(f) represents the Fourier transform of the current time segment of the said observation signal.

Images