FR2888704A1 - - Google Patents

Abstract

Disclosed is a method for discriminating and attenuating echoes of a multi-layered hierarchical encoded digital audio signal from transform coding, echo generator, and predictive, non-echo generator coding, comprising: at least at decoding, for each current frame of the digital audio signal, to compare (A), in real time, the value of the ratio of a representation of the temporal envelope of the signal resulting from an echo generator decoding to that of the signal from non-echo generator decoding; and if the value of said ratio is greater than or equal to a threshold value, concluding (B) the existence of an echo from the transform coding in the current frame; otherwise, if the value of the ratio is smaller than said threshold value, to conclude (C) that there is no echo resulting from transform coding in the current frame, to be attenuated (D), conditionally to the existence of echo, the signals from the echo generator transform decoding to reduce the audibility of the echoes.Application to the implementation of echo sound attenuation devices of a digital audio signal and hierarchical coders incorporating such devices.

Description

METHOD AND DEVICE FOR MITIGATING THE ECHO OF A SIGNAL

  AUDIONUMERIC ISSUANCE FROM A MULTILAYER ENCODER The digitization of the source signals into a digital audio signal requires a higher rate of flow as the quality of the returned signal is high.

  To adapt the transmission rate of the above-mentioned digital audio signals, compression processes are used by virtue of the implementation of coding systems of the CELP coding or transformation coding type.

  Encoders in the CELP coding family are based on synthetic analysis predictive coding techniques for bit rates in the range of 4 to 32 kbps, essentially in the field of speech coding.

  The signal is divided into 10 to 30 ms slices on which the representative parameters of the latter are coded. The aforementioned CELP coders induce, on the obtained digital audio signal, essentially degradations of the type of granularity effect resulting from the quantization noise, in the absence of significant pre- or post-echo phenomenon.

  The coders of the transform coding family perform a time-frequency transposition of the signal in a transformed domain, Fourier, MDCT or other, a quantization of the transformed domain coefficients taking into account a psycho-acoustic model, and then a transmission of the coded coefficients. to a decoder in the form of a binary frame.

  Used for a wide variety of audible signals at 24 to 64 kbit / s per channel, audible signals with strong attacks or decays are the seat, on decoding, of noise from quantization in the form of pre- or post-echoes.

  The aforementioned encoders can operate at a fixed or variable rate, the flow rate being adaptable.

  Some of the aforementioned coders, designated hierarchical coders, finally have the capacity to produce a binary train, said nested or hierarchical, whose least significant bits can be ignored while maintaining a correct coding quality. The bit frame is then structured by ordering the bits from the essential bits to the least significant bits.

  The above-mentioned coders can advantageously be used for personal communications on the INTERNET, for example where the bit rate available for transmission can decrease significantly at one point in the network.

  During transmission, it is then possible to ignore the least significant bits, while keeping a significant bit stream. The quality of signal reproduction is a function of the number of decoded bits.

  The object of the present invention is to reduce the audibility of echo noise inherent in audio compression systems having a plurality of coding layers, such as hierarchical coders, at least one of which is not the seat of the encoder. pre- or post-echo phenomenon.

  Prior to the presentation of the invention, a reminder of the principle of encodingecoding by transform will be introduced hereinafter, with reference to FIG.

  Figure 1 depicts an example of transposition encoding in the frequency domain and corresponding decoding. In this example, in coding, the transposition of a time domain signal block in the frequency domain is performed by an MDCT transform applied to the windowed signal by a sinusoidal function which makes it possible both to increase the frequency resolution and to satisfy the conditions perfect reconstruction of an MDCT overlay analysis-filter bench. Other types of windows can be used. Quantification of spectral coefficients uses different methods such as quantization by a synthesis analysis method or an interactive quantization algorithm. These methods are known in the state of the art and will not be described in detail for this reason.

  In the aforementioned quantization methods, the quantization operation of the spectral coefficients is performed taking into account a minimization constraint of the audible quantization noise, by taking into account a psychoacoustic model. Quantification is followed by Hufman or other entropic coding, which takes into account the non-uniform probability densities of the spectral coefficients.

  At decoding, the spectral coefficients are decoded by an entropic decoder, dequantized and then transposed into the time domain by inverse MDCT transform, denoted MDCT INV in FIG. 1. An overlay processing, "add-overlap" in English, allows obtain the reconstituted signal from two successive partial recovery frames, current frame of rank m and previous frame, of rank m-1.

  The reconstituted signal is of the form: xrec (n) = h (n + 2) xprev (n + 2) + h (n) x, r (n) for n [0, 2 -1].

  In the preceding relation, in the range of the samples of rank between 0 and N / 2-1: xc r (n) designates the sample of the current frame of rank m; h (n) denotes the window of the inverse MDCT transform applied to the current frame; - xprev (n + Y) denotes the sample of the previous frame of rank m-1, sample shifted by N / 2 to ensure recovery; h (n + 2) denotes the window of the inverse MDCT transform applied to the previous frame, window shifted by N / 2 to ensure recovery; xrec (n) is the time-domain reconstituted signal sample.

  Certain musical sounds, such as percussion, and for certain sequences of words, especially plosive syllables, are characterized by extremely abrupt attacks that result in very rapid and wide-ranging transitions of sound signals in the time domain.

  In this situation, the sampling of the sound signal results in a very strong variation in the dynamics of the signal sampled over a few samples.

  The cutting into successive blocks of samples carried out by the transform coding is totally independent of the sound signal and the transitions therefore appear at any point in the analysis window. In transform coding, the noise is distributed temporally evenly over the entire duration of the sample block. This results in the appearance of pre-echoes prior to the transition and post-echoes after the transition.

  The noise level is lower than that of the signal for the high energy samples, immediately following the transition, but it is higher than that of the signal for the lower energy samples, especially on the part preceding the transition. For the aforementioned part, the signal-to-noise ratio is largely negative and the resulting degradation, designated pre-echoes, can appear very troublesome.

  Indeed, the human ear performs a fairly limited pre-masking, of the order of a few milliseconds, before the physiological transmission of the attack.

  The noise produced, or pre-echo, is audible when the duration of the pre-echo is greater than the duration of the pre-masking.

  The human ear also performs a post-masking of a longer duration, from 5 to 60 milliseconds, during the passage of high energy sequences to low energy sequences. The rate or level of discomfort acceptable for post-echoes is therefore greater than for pre-echoes.

  The phenomenon of pre-echoes, more critical, is even more troublesome as the length of the blocks in number of samples is important. However, in transform coding, it is necessary to have a good frequency resolution to allow the allocation of bits on the relevant areas of a signal spectrum, hence the advantage of using very long blocks. . When a coding process, AAC (Advanced Audio Coding) for example, is implemented, a window of great length contains a fixed number of samples, 2048, or over a period of 64 ms at a frequency of sampling of 32 kHz. Encoders used for conversational applications use a window of 40 ms duration and a frame renewal duration of 20 ms.

  To reduce the aforementioned annoying effect of the pre-echo phenomenon, and to a lesser extent post-echo, various solutions have heretofore been proposed.

  A first solution is to apply filtering. In the zone preceding the transmission due to the attack, the reconstituted signal consists in fact of the original signal and the quantization noise superimposed on the signal.

  A corresponding filtering technique has been described in the article High Quality Audio Transform Coding at 64 kbits, IEEE Trans on Communications Vol 42 No. 11, November 1994, published by Y. Mahieux and J. P. Petit.

  By such a process, when a sample block is assigned pre-taps, filtering operating at the decoder on the reconstructed samples after addition / overlap reduces the power of the spurious signal. Based on the Kalman filtering principle, the aforementioned filtering is an adaptive filtering so as to follow any non-stationarity of the signal in the part of the sample block preceding the attack. Kalman filtering is applied to the samples located before the transmission and delivers the denoised signal which replaces, for the corresponding part of the block, the samples resulting from the addition / recovery.

  The implementation of such a filtering requires the knowledge of parameters some of which are estimated at the decoder from the noisy samples. On the other hand, information such as the energy of the original signal can only be known to the encoder and must therefore be transmitted. When the received block contains a sudden variation of dynamics, the filtering treatment is applied to it.

  The aforementioned filtering process does not allow to find the original signal, but provides a strong reduction of pre-echoes.

  A second solution is to reduce pre-echoes by a dynamic concentration of windows.

  Such a technique has been described by US Pat. No. 5,214,742 issued to B. Edler. This solution has been the subject of applications in different audio coding solutions according to international standards.

  According to this solution, due to the fact that the temporal and frequency resolution of the signals strongly depends on the length of the coding window, the frequency coders switch between long windows, of 2048 samples, for the stationary signals, and short windows 256 samples for signals with large variation of dynamics or transients. This adaptation is performed in the AAC module, the decision being taken frame by frame to the encoder. In fact, to preserve the perfect reconstruction property, a sequence of short windows is preceded and followed by a transient window. The selection is valid for the second part of the window only because the shape of the first part depends on the decision made at the previous frame.

  One of the disadvantages of this second solution is that it induces an additional delay of the order of N / 2 samples due to the fact that if a transition begins in the next window, it must be possible to prepare the transition and to switch to a transition window to keep the perfect reconstruction.

  The reduction of echoes can, however, be facilitated when the decoder consists of several stages of transformation, as is the case for hierarchical coders. In this case, the predictive coding stages which are often core core or core layer can be used for the detection of the echo. An example of decoding of this type is described in US Patent Application No. 2003/0154074 to K. Kikuiri et al. The procedure described by the aforementioned patent application consists in detecting pre-echoes exclusively from the decoded base CELP signal.

  Such a procedure does not ensure, for this reason, a joint and synchronous pre-echo reduction processing of the reconstructed frames.

  The present invention relates to the implementation of a discrimination method and a device for discriminating and attenuating the echoes of a digital audio signal from a multilayer or hierarchical encoder.

  The method for discriminating the echoes of a digital audio signal generated by multilayer hierarchical coding from a transform coding, an echo generator, and a non-echo generator predictive coding object of the invention is remarkable in that it consists at least, in the decoding, for each current frame of the digital audio signal, to compare, in real time, the value of the ratio of the amplitude of the signal resulting from an echo generator decoding to the amplitude of the signal from a non-echo-decoding decode to a threshold value. If the value of this ratio is greater than or equal to this threshold value, it consists in concluding that there is an echo resulting from transform coding in the current frame. Otherwise, the value of this ratio being lower than this threshold value, it consists in concluding that there is no echo resulting from transform coding in this current frame.

  The device for discriminating and attenuating the echoes of a digital audio signal, which is the subject of the invention, operates on such a signal generated by a multilayer hierarchical coder, this encoder comprising a first plurality of predictive, non-echo-generating decoders. and a second plurality of transform decoders capable of revealing echoes. Each predictive encoder has a first output providing a decoded digital audio signal and a second output providing an auxiliary signal representative of the decoded digital audio signal energy. Each decoder per transform delivers, after decoding by inverse transform and overlap addition, a decoded signal succession of current frames.

  It is remarkable in that it comprises at least at the level of a decoder, a module for calculating the existence, the origin position of the echo and the echo attenuation value applicable in the decoder. current frame receiving, on the one hand, the auxiliary signal delivered by the second output of a predictive decoder of rank i of this plurality of predictive decoders, and, on the other hand, the decoded signal delivered by the output of a decoder by inverse transform of rank j of this plurality of decoders by inverse transform, this calculation module delivering a presence signal, echo position and echo attenuation value applicable in the current frame and an attenuation module of the echo receiving the decoded current frame succession signal delivered by the row-j inverse transform decoder and the presence, echo origin position and echo attenuation value applicable signal in the current frame and delivering a decoded signal succession of current frames in which the echo is attenuated from the original echo position.

  The method and the device, objects of the invention, are applicable to the processing of sound signals, in particular coded digital audio signals whose frames are the seat of transition of attack and / or decline generated by musical instruments, voice signals comprising plosive syllables, and, more generally, multilayer encoder and decoder devices including predictive coders and frequency transform coders respectively of predictive decoders and inverse frequency transform decoders.

  They will be better understood on reading the description and on the observation of the drawings below, in which, in addition to FIG. 1 relating to the prior art: FIG. 2a represents, by way of illustration, an experimental chronogram of the signals coming from a hierarchical multilayer coder making it possible to justify the implementation of the method which is the subject of the invention; FIG. 2b represents, by way of illustration, a flowchart of the essential steps of the method that is the subject of the present invention implemented during decoding, based on the justifications made with reference to FIG. 2a; FIG. 2c represents, by way of illustration, a flowchart of the essential steps of the method that is the subject of the present invention represented in FIG. 2b, made conditional on the detection of the existence of echoes in a current frame, a signaling of this existence being transmitted to the decode together with the current frame; FIG. 3a illustrates the general diagram of a multilayer hierarchical decoder; FIG. 3b represents, by way of illustration, a functional block diagram of an echo discrimination and attenuation device, object of the invention, implanted in a multilayer hierarchical decoder; FIG. 3c represents, by way of illustration, an implementation detail z.sub.q of a module for calculating an existence discrimination signal, an echo origin position signal, and an attenuation value signal applicable to a current frame incorporated in the device object of the invention shown in Figure 3b; FIG. 3d represents a functional flowchart for calculating the echo existence and the origin position of the echo in the current frame implemented in the calculation module represented in FIG. 3c; FIG. 3e represents a functional flowchart for calculating the attenuation value applicable to the current frame implemented in the calculation module represented in FIG. 3c; FIG. 3f represents a functional flowchart for calculating the existence of echo and the original position of the echo in the current frame implemented in the calculation module represented in FIG. 3c, this computation process being rendered conditional on the detection at the coding of the existence of echoes in the current frame, in accordance with the method which is the subject of the invention; FIG. 4a represents a first alternative embodiment of the device that is the subject of the invention, the attenuation of the echo being applied directly at the output of the inverse transform decoding module, before the recovery addition process; FIG. 4b represents, by way of illustration, the overlap of a current frame seated with a transition SI and a preceding frame, seat of pre-echoes; FIG. 4c represents a timing diagram of the variation of the signal envelope ratio, taking into account the existence of the transition shown in FIG. 4b; FIG. 4d represents, by way of illustration, a detailed implementation of a module for calculating an existence and position discrimination signal of an echo incorporated in the device of the invention represented in FIG. 4a. ; FIG. 4e represents, by way of illustration, a functional flowchart for calculating the attenuation value applicable to the current frame implemented in the calculation module represented in FIG. 4a, taking into account discrimination of the pre- and post- echoes; FIG. 4f represents a detail of implementation of an attenuation module, by way of non-limiting example; FIG. 5a represents, by way of illustration, a second alternative embodiment of the device of the invention comparable to that of FIG. 4a, in which the implementation of an auxiliary overlay add module is suppressed in a purpose of simplification and cost reduction; FIG. 5b represents, by way of illustration, a detail of implementation of a module for calculating an existence discrimination signal, an echo origin position signal and an attenuation value signal applicable to a frame. current more particularly adapted to the module 1 of the second embodiment of the device object of the invention shown in Figure 5a; FIG. 6a and FIG. 6b represent, by way of illustration, modules for calculating an echo presence signaling message in a current frame that can be implanted in a multilayer hierarchical type coder, in order to enable the method to be implemented. and the device objects of the invention; FIG. 7 represents, by way of illustration, an alternative embodiment of FIG. 3b, in which the signal originating from the predictive decoder of rank i is transmitted to the decoder by rank transformation j in the frequency domain; FIG. 8 represents, by way of illustration, an alternative embodiment of FIG. 4a, in which the signal coming from the predictive decoder of rank i is transmitted to the decoder by rank transformation j in the frequency domain; FIG. 9 represents, by way of illustration, an alternative embodiment of FIG. 5a, in which the signal originating from the predictive decoder of rank i is transmitted to the decoder by rank transformation j in the frequency domain; FIG. 10 represents, by way of illustration, an improvement to the modes of implementation of FIGS. 3b and 7, whereby echo detection and attenuation are performed on a first channel in a low frequency band and on a least another channel in separate frequency bands.

  FIG. 11 represents, by way of illustration, an improvement in the embodiment of FIG. 10, whereby the detection and the attenuation of the echo are performed on a first channel in a low frequency band and on a least another channel in a separate frequency band, the predictive signal being filtered into a plurality of subbands.

  A description of the echo discrimination method of a digital audio signal generated by multilayer hierarchical coding from transform coding, echo generator, and predictive, non-echo generator encoding, in accordance with The object of the present invention will now be given with reference to FIGS. 2a and 2b.

  FIG. 2a represents a recording (amplitude, number of samples) on a frame of a first curve 1 representative of the calculated ratio to the coding between the difference signal between the delayed original signal and the coded and decoded signal by a non-predictive coding. echo generator (residual signal), vis-à-vis the output signal delivered by the predictive decoder.

  The second curve 2 represents the ratio between the aforementioned difference signal encoded and decoded by echo generator transform coding and the output after decoding by a predictive decoder (for example a CELP decoder).

  The above-mentioned curves 1 and 2 make it possible to deduce that the ratio is much greater in the presence of pre-echoes.

  The experimental observation shown in FIG. 2a makes it possible to justify the implementation of the method that is the subject of the invention, as represented in FIG. 2b.

  The method according to the invention makes it possible to discriminate the echoes of a digital audio signal during decoding, when this digital audio signal is generated by multilayer hierarchical coding from a transform coding and a predictive coding.

  With reference to FIG. 2b, the signal delivered by an inverse transform decoding delivered by a layer transform decoder of a multilayer hierarchical decoder is denoted by: x. x,% (n) the signal delivered by a predictive decoding executed by a predictive decoder of layer i in the corresponding hierarchical decoder. The signal 4 (n) may be either the output signal of the non-echo generating predictive decoder or a filtered version of this signal or a representation of the short-term energy of this signal.

  With reference to FIG. 2a and FIG. 2b, it is indicated that the method which is the subject of the invention consists of a step A to compare in real time the value of the ratio of the amplitude of the signal resulting from a generator decoding. echoes the amplitude of the signal from a non-echo generator decoding to a threshold value S. In FIG. 2b it is indicated that the amplitude of the signal resulting from an echo generator decoding is denoted EnvTj (k) and the amplitude of the signal from a non-echo-generating decod and denoted Envp, (k).

  With reference to the notation indicated, it is understood in particular that the amplitude of the signal resulting from an echo generator decoding and the amplitude of the signal coming from a non-echo generator decoding can advantageously be represented by the envelope signal. of the echo generator decoding signal x,., (n) respectively of the signal resulting from a non-echo generator decoding xP1 (n). In FIG. 2b, the obtaining of the amplitude signal is represented by the relations: xTi (n) - + EnvTj (k) xP; (n) - + Envp, (k).

  In general, it is indicated that the amplitude signal signal from an echo generator decoding respectively the signal from a non-echo generator decoding can be represented not only by the aforementioned envelope signal but also by any signal such as the absolute value, or other, representative of the aforementioned amplitude.

  With reference to the same FIG. 2b, it is indicated that the ratio of the amplitude of the signal resulting from an echo generator decoding to the amplitude of the signal coming from the non-echo generator decoding is represented by the relation: EnvT (k ) R (k) = Envp; (k) - 13 - Referring to the preceding notations, it is indicated that the comparison step A of FIG. 2b consists of comparing the value of the ratio R (k) with the threshold value S by comparison of superiority and equality .

  If the value of the aforementioned ratio is greater than or equal to the threshold value S, in positive response to step A, the aforementioned test then makes it possible to conclude in step B that an echo resulting from the coding by transformed in the current frame, this echo being then revealed at the decoding.

  The existence of the echo is represented in step B by the relation: 3 echo xri (n). Otherwise, in a negative response to the test of step A, if the value of the aforementioned ratio is less than the threshold value S, the test of step A then makes it possible to conclude in step C that there is no echo from the transform coding in the current frame.

  This relation is noted in step C by: ZI echo x7i (n). In a particularly advantageous manner, in accordance with the implementation of the method which is the subject of the invention, it is indicated that the original position of the echo in the current frame is actually given by the position, in the current frame, the value of the ratio substantially equal to the threshold value S. The value mentioned above is given in step B of FIG. 2b by the relation: Pos = k1R (k) = S.

  In general, with regard to the implementation of the test of step A and finally tests C and B of FIG. 2b, in particular of step B following step A, it will be understood that the calculation of the value of the ratio R (k) can be carried out as a smoothed value on the current frame, so as to compare in real time the value of the aforementioned ratio with the threshold value S. When the value of the aforementioned ratio is equal at the value of S, then the original position of the echo is given by the particular value of the rank k of the corresponding sample of the signal decoded in the current frame. The above calculation process will be described in such a way

  more detailed later in the description. The method which is the subject of the invention can moreover be implemented by means of

  non-limiting manner by preprocessing performed coding on each current frame so as to detect the existence of an echo in the current frame considered. Such an implementation is shown in Figure 2c and will be described with reference to the aforementioned figure.

  FIG. 2c denotes by C x,) (n) and C 4 (n) the digital audio signal coded from a transform coding, an echo generator, respectively from a predictive coding, which is not a generator. echo, this signal being transmitted in the form of successive current frames, as mentioned previously in the description.

  Thus, the method, which is the subject of the present invention, consists in coding for each transmitted frame, in a step OA, to perform the detection of the existence of echo in the current frame, the detection of the existence of echoes being able to of course, be implemented according to the operating process of Figure 2a, as will be described later in the description. This operation corresponds to the implementation of step OA of FIG. 2c and makes it possible to generate an echo presence variable Ec which can consist of a binary variable of value 1 or 0 constituting in fact a flag representative of the existence or non-existence of echoes.

  The step OA is then followed, as represented in FIG. 2c, by a joint transmission step OB, of decoding coding, of the current frame, that is to say signals coded C (n) and C xp; (n) and the variable Ec as an echo existence signaling message.

  At decoding, as shown in FIG. 2c, the decoding operations of the received frames are of course carried out in order to obtain the decoded signals, that is to say the signals xi) (n) and xP; (n) according to the relations: C xT (n) - * xi, (n) C xp; (n) xP (n) and of course the calculation of the ratio R (k) as described previously with reference to FIG. 2b.

  The aforementioned decoding step then makes it possible to perform echo discrimination on decoding conditionally on the existence or non-existence, a priori, of discriminated echoes in the encoding.

  Thus, following the actual decoding operation and the calculation of the ratio R (k), the method, which is the subject of the invention, can consist, in a step Ao executed at decoding, in verifying the true value of the variable Ec echo existence a priori.

  On negative response to test Ao of FIG. 2c, that is to say when the current frame has no prior echo, step C of absence of echo in the current frame as represented in FIG. 2a is called in Figure 2c.

  On the contrary, on a positive response to the Ao test, that is to say in the presence of a priori echo, the steps A and B as represented in FIG. 2b are then implemented in a similar manner.

  Preferably, and according to a remarkable aspect of the method, object of the invention, the echo existence signaling message is a false alarm message M (Ec) echo presence in the current frame. Under these conditions, the Ec echo signaling variable Ec may advantageously take the value 0 when there is an echo in the current frame and, conversely, the value 1 when there is no echo in the echo. the current frame. The aforementioned values are of course given by way of non-limiting example, taking into account the presence discrimination respectively absence of echo in negative logic of false alarm.

  In general, it is recalled that the notion of echo detection in the current frame coding relates more particularly the frame information with or without false echo, the echo being simply revealed at decoding. Indeed, there are no echoes to the coding itself, since the calculations are performed on the unquantized signals. This implies that if the echo is detected at the encoder and is also detected at the decoder it is in fact a false alarm. It is then incorrectly detected at the decoder and the echo control procedure should not be applied.

  The method, object of the invention, then advantageously consists in transmitting the false alarm message as described previously in the description.

  The implementation of the method, object of the invention, with calculation and establishment of a false alarm message, can then be implemented in the encoder in specific cases and only for certain configurations of CELP coders used in the first layer of hierarchical coders. and coding a restricted signal band between 0 and 4 kHz for example.

  A more detailed description of a device for discriminating and attenuating echoes of a digital audio signal generated by a multilayer hierarchical coder, in accordance with the subject of the present invention, will now be given in connection with FIG. 3a and FIGS. following.

  FIG. 3a shows a multilayer hierarchical coder comprising predictive decoder decoding stages, denoted layer 1 to layer i, and transform decoder decoding stages, denoted layer 1 to layer j. Each of the aforementioned decoders respectively delivers a decoded audio signal denoted xpi (n) to xp, (n) and xn (n) to x7 (n) respectively.

  In addition, the predictive decoders of layer 1 to i comprise a second auxiliary output delivering an auxiliary signal denoted x, 1 (n) to 4 (n). The decoded output signals of the predictive decoders and the transform decoders are delivered to a somer So which delivers the reconstructed decoded signal xs (n).

  The aforementioned auxiliary signal delivered by each of the predictive decoders is either the output signal of the non-echo-generating predictive decoder or a filtered version of this signal or an auxiliary signal representative of the energy of the decoded digital audio signal, this auxiliary signal being able to be the output signal of the CELP decoder, the excitation signal of the CELP decoder or the envelope of these signals, or finally any other signal representative of the energy of the decoded predictive signal delivered by the predictive coder.

  Each decoder per transform delivers, after decoding by inverse transform and overlap addition, a decoded signal succession of current frames. In some configurations this signal can then be filtered by an inverse perceptual filter 1 / W (z). This filter is necessary in configurations where the transform coder encodes a version of the filtered signal by a perceptual filter W '(z). The coefficients of this filter are extracted from the received bitstream. These filters are shown in dashed lines in FIG. 3a. In all the cases discussed below, the signal decoded by a transform decoder in which the echo is attenuated will possibly be followed by such inverse perceptual filtering, especially in the case where perceptual filtering has been introduced into the encoder. In the following, these filters will no longer be mentioned in order to lighten the figures.

  Each transform decoder receives the aforementioned auxiliary signal.

  With reference to FIG. 3b, the device that is the subject of the invention then has, in a particularly advantageous manner, a module 1 for calculating the existence, the original position of the echo and the applicable attenuation value. in the current frame, this module receiving, on the one hand, the auxiliary signal xP; (n) delivered by the second output of the predictive decoder of rank i of the plurality of predictive decoder and, on the other hand, the decoded signal x, j (n) delivered by the output of a decoder by inverse transform of rank j of the plurality of decoders by inverse transform. The calculation module 1 above provides a presence signal, echo position and echo attenuation value applicable in the current frame, this signal being noted DPG in Figure 3b.

  In addition, in order to ensure the attenuation of undesirable echoes, the device which is the subject of the invention comprises an echo attenuation module 2 receiving the decoded signal common frame succession delivered by the inverse transform decoder. rank j and, of course, the presence signal, the echo origin position and the echo attenuation value applicable in the current frame, that is to say the DPG signal delivered by the module 1 .

  In a nonlimiting manner, the echo attenuation module 2 is represented in FIG. 3b connected at the output of the floor covering addition module j, which is itself connected at the output of the spectral coefficients decoding module. by inverse transform of corresponding rank. The echo attenuation module 2 delivers a decoded current frame succession signal in which the echo is attenuated or substantially suppressed, from the original position of the echo. This signal is of course delivered to the somer So shown in Figure 3b, which plays the same role as in the case of Figure 3a.

  It is understood, in particular, that the module 1 for calculating the existence and the original position and the attenuation value applicable in the current frame makes it possible to deliver the signal DPG in accordance with the implementation of the object process of the invention as represented in FIG. 2b when the ratio of the envelopes is greater than the value of the threshold S, the attenuation value of the echo applicable in the current frame being further communicated to the attenuation module 2 of echo, as will be described hereinafter with reference to FIG. 3c.

  FIG. 3c represents a particular nonlimiting embodiment of the module 1 for calculating the existence of the echo origin position as well as the echo attenuation value applicable in the current frame.

  As represented in FIG. 3c above, the module 1 advantageously comprises a module 1 o for calculating the time envelope of the decoded signals xa, (n) and x,., (N) delivered by the respective stages i and j of the decoders predictive or inverse transform decoders, respectively.

  The module 1 further comprises a module 11 for calculating the envelope ratio from the signals Envp, (k) and Env, i (k) delivered by the temporal envelope calculation module 10.

  The module 11 then delivers the value R (k) of the aforementioned ratio to a decision and position calculation module 12 and the attenuation value to finally deliver the aforementioned DPG signal.

  The temporal envelope calculation module 1 o can advantageously comprise a first specific calculation module that makes it possible to calculate the envelope of the auxiliary signal 4 (n) according to the relation: N 1 Env p, (k) = xP1 ( kN + n) 2, k = 0, ..., K 1. = o

  In the previous relation: n is the rank of the sample, N is the length of the sub-block on which is calculated an envelope point, K is the total number of sub-blocks in a frame of length K N. the first specific module may also be constituted by a digital module for calculating the envelope of the aforementioned auxiliary signal xP1 (n) Envp, (n) = aEMaxuxP1Envp, (n 1)) + (1 a) ÉIxP, a (n.

  In addition, the module 10 may comprise a second specific module for calculating the envelope of the output signal of the inverse transform decoding module according to the equation: N 1 Envr, (k) = xTJ (kN + n) 2, k = 0, ..., K 1. n = o

  In the previous relation, n, N and K designate the same elements as in the previous relation.

  The second specific module can also be constituted by a digital module for calculating the envelope of the aforementioned signal according to the relationship for the decoded signal by inverse transform: Env, i (n) = aEmax (JxTi (nl, EnvT, (n 1)) + (1 a) ÉIxTj (n).

  In the above-mentioned corresponding calculation mode, the envelope is in fact calculated by performing a first-order low-pass filtering on the absolute value of the decoded signal.

  The two modes of calculating the envelope value of the decoded signal by inverse transform are, of course, given by way of non-limiting example.

  The module 11 for calculating the envelope ratio is now described when the envelope is calculated by sub-blocks of N samples.

  The above-mentioned module 1 1 then advantageously comprises a specific calculation module directly delivering the ratio R (k) according to the following relationship: Envi. (k) R (k) = k = O, K 1. EnvP; (k) In addition, in a nonlimiting preferred mode of implementation, a smoothed envelope ratio calculation can be carried out by means of a specific module given in each successive sample by the value Env, (n) Rrs (n) = n = 0, ..., KN -1. eNVP; (n) It is understood that the smoothed envelope ratio is then calculated on a frame of length K N substantially.

  Finally, the module 1 comprises a module 12 for calculating echo presence decision, echo origin position and attenuation value applicable to the current frame, by comparing the value of the ratio R (k) of the values. The module 12 delivers the signal DPG under the following conditions, which will be explained and described in connection with Figures 3d, 3e and 3f.

  FIG. 3d represents a flowchart of the essential steps implemented by an echo presence decision calculating module and determining the origin position of the echo.

  The aforementioned module may consist of a software module which, as shown in FIG. 3d, comprises an initialization step 100 consisting of initializing the rank k at the value 0, rank of the sample of the current frame by

example.

  Step 100 is followed by a step 101 of comparing the value of the envelope ratio R (k) with the threshold value S by comparison of superiority or equality.

  On a negative response in step 101, a step 102 is called comparing the rank of the sample k to the maximum value K-1 by comparison of strict inferiority. The positive response to the above-mentioned test 102 makes it possible, in a step 103, to increment the value k of the rank of the sample to the value k + 1 for a return to the test 101 for comparing the value of the ratio with the value of threshold S. - 21 - When on the contrary, in step 102, the answer is negative, that is to say when the computation, on the one hand, of the value of the ratio R (k) and of comparison on the other hand, from this value to the threshold value S has been made for all the samples, a step 105 is called which consists of noting the absence of echo and returning to an end step 106.

  On the contrary, on a positive response to the test 101, a step 104 and called which is to note the presence of echo and to instantiate the position value of the origin of the echo, noted value Pos, to the corresponding value k of the rank of the sub-block, the original position of the echo corresponding in this case to the rank of the sample considered.

  Step 104 is then followed by step 106 of end of echo discrimination.

  An example of implementation of a specific calculation module constituted for example by a digital software calculation module of the attenuation value applicable to the current frame to perform the suppression or attenuation of unwanted echoes is shown in FIG. .

  The above-mentioned software calculation module can be implemented as previously described in the module 12 shown in FIG. 3e or, if appropriate, in the attenuation module 2 itself described previously in the description in conjunction with FIG. of the scope of the subject of the present invention.

  With reference to FIG. 3e, the position of the echo in step 200 k = Pos. This position has been determined previously as described in connection with FIG. 3d.

  Step 200 is followed by a step 201 of calculating the attenuation value, that is to say finally an applicable gain for each sample from the original position of the echo .

  According to a remarkable aspect of implementation of the method and the device which is the subject of the invention, it is indicated that the attenuation value is either the value 1 in the absence of the existence of an echo, or the value of the inverse the ratio of temporal envelope values, in the presence of echo, according to the relation: g (k) min (EnvPj (k), l.

  EnvTi (k) The value finally retained corresponds to the minimum value between the value 1 and the value of the inverse of the ratio of the aforementioned temporal envelope values.

  It is thus understood that, in the absence of an echo, the value 1 is applied but that in the presence of echo, the application of an attenuation value, or gain value, then makes it possible to restore a situation comparable to that of the absence of echo, due to the application of the inverse value of the ratio of the envelope signals.

  It is thus understood that the method and the device which are the subject of the invention appear particularly remarkable insofar as the echo attenuation corresponds substantially to an automatic gain correction applied as a function of and from the original position. echoes.

  Step 201 is in fact followed by a step 202 which consists in initializing the value of n rank of the sample at the value 0, ie the value of the effective sample rank from which the attenuation is applied in the current frame.

  Step 202 is followed by step 203 which consists in actually applying the value of the attenuation g (k) to the input signal Si (k N + n) to generate the output signal So 1 (k N + n) according to the relation: So (k N + n) = Cte g (k) S ,,, (k N + n). Step 203 is then followed by step 204 of comparing the value of n by inferiority comparison to the N-1 value, which is in fact the rank of the penultimate sample of the current frame.

  On a positive response in step 204 a step 205 is called to increment the rank of the sample n to the value n + 1 followed by a return to the application of the attenuation value g (k) to the Step 203. This process is continued as long as the rank n of the sample is less than N-1.

  On the contrary, on a negative response at step 204, the last sample of the current frame having been processed, a step 206 is called, which consists in comparing the value of the rank k of the sample with the value K- 1 second to last value of the rank of the sub-blocks in a frame of length K N. On positive response to the test 206, a step 207 is called consisting in incrementing the value of k to the value of k + 1 and then returning to the step 201 for calling the attenuation value g (k) and continuing the process as long as the rank of the last block is not reached.

  On the contrary, on a negative response to the test 206, the last sub-block having been reached, an end step 208 is called, the current frame having thus been processed and the echoes attenuated on the latter.

  It will be understood, in particular, that with regard to the implementation of FIG. 3e, this is advantageously performed at the attenuation module 2 represented in FIG. 3b, but that, if necessary, the value of FIG. attenuation g (k) as described in block 201, can be calculated in the module 12 shown in Figure 3c and transmitted directly to the attenuation module 2 for example. Such a mode of implementation is not limiting and does not go beyond the scope of the subject of the present invention.

  Finally, as shown in FIG. 3b, the device that is the subject of the invention further comprises, at the level of the decoder, a reception and processing circuit of the echo-existence signaling message transmitted together with each current frame of the encoder. to the decoder.

  In the above-mentioned FIG. 3b, the reception and processing circuit is represented in dotted lines denoted by Ec and is deemed to allow the processing and reading of the value of the variable Ec representative of the presence or absence of echo in the frame. current transmitted. Such a circuit may correspond to the reading of the value of a flag bit in the false alarm signaling message for example.

  Such a circuit is known as such and, for this reason, will not be described in detail.

  As further represented in FIG. 3f, for a signaling message of the false alarm message type of presence of echoes in the current frame, the device according to the invention comprises associated with the positional decision calculation module 1 echo origin and the attenuation value applicable to the echoes in the current frame, a software module for reading the value of the variable Ec, that is to say the false alarm message.

  On reading in step 300 of FIG. 3f of the false alarm message and of the variable Ec at the true value, the information delivered by the message corresponds to an a priori echo absence on the current frame, it is ie Ec = 1 for example, that is to say on negative response to the test 300, the echo presence is set to 0 in step 301, ECHO = O by inhibiting the decision calculation module of existence of echo origin position and applicable attenuation value, none of the steps 100, 101 to 106 of Figure 3d and corresponding to Figure 3f being then used to determine the position of echo. Step 301 is then followed by a call step of step 106 of end of determination of the echo position. It will be understood that step 301 performs an inhibition of the echo origin position and applicable attenuation value decision calculation process.

  On the contrary, on a positive response to the test 300 of FIG. 3f, the information delivered by the message and in particular by the variable Ec, Ec = 0 in the example given, corresponds to a presence a priori of echo on the frame common.

  In this situation, the positive response to the above-mentioned test 300 makes it possible to activate the decision-making process of existence, echo origin position and attenuation value applicable according to steps 100 to 106 of FIG. 3d. and corresponding FIG. 3f. It is understood, in particular, that the method, and the device objects of the invention, when these are implemented in accordance with FIG. 2c and FIG. 3f, allow a saving in calculation cost at decoding taking into account the the inhibition of the original position echo and attenuation value presence calculation process implemented, thanks to the a priori determination of the existence of an echo in the current frame at the coding and the encoder level original.

  In FIG. 3b, the dotted arrow from the predictive coding part to the decoding part of the spectral coefficients represents the possibility of reconstituting the spectral coefficients by partially using the predictive decoder output. By way of nonlimiting example, a solution consists in performing a time-frequency transformation of the output signal of the predictive decoder and then in subtracting decoded spectral coefficients, the spectral components of the predictive signal which it is desired to eliminate from such a fate that the reconstituted time signal XS (n) can be obtained by adding the output signal of the predictive decoder and the output signal of the frequency-time transformation after attenuation. Indeed, this procedure is effective because the received transform coefficients represent the signal and they can advantageously replace the spectral coefficients of the predictive signal in certain frequency areas.

  The pseudo code for performing the operation "spectral coefficient decoding" of Figure 3b is given below.

  Pseudo Code for k = 0, ..., Kb-1 CoeffRi (k) = CoeffTj (k) end for k = Kb,. .., Kf-1 if (flag (k) == dispo) CoeffRi (k) = CoeffTj (k) - Coeffp, (k) else CoeffRR (k) = 0 endif end The coefficients Coeffp; (k) are obtained by time-frequency transformation of the signal from the predictive decoder of the layer i. This operation is performed in the "spectral coefficients decoding" box of FIG. 3b.

  A first variant embodiment of the device that is the subject of the invention will now be described with reference to FIG. 4a and the following figures.

  The above implementation variant is implemented for a reverse-conversion decoder of rank j comprising an inverse frequency decoding module designated decoding spectral coefficients inverse MDCT stage in the aforementioned figure connected in cascade with a recovery addition module.

  However, as shown in FIG. 4a, the device that is the subject of the invention comprises an echo attenuation module 2 interposed between the inverse decoding module of the spectral coefficients and the overlay adding module.

  In addition, an auxiliary overlap adding module with the reference 3 is provided, which receives two successive frames of decoded spectral coefficients delivered by the inverse decoding module of the layer j and delivers an auxiliary decoded signal which is none other than the signal (n) previously described in the description. In the aforementioned signal corresponding to the current frame, the echoes are not attenuated. This signal is delivered to module 1 for calculating the existence of the original position of the echo and the attenuation value in the current frame. The module 1 for calculating the existence of the original position of the echo and the attenuation value applicable in the current frame then delivers the presence signal of echo position and attenuation value DPG in the current frame at the echo attenuation module 2.

  The mode of implementation of the device according to the invention, as represented in FIG. 4a, involves a direct operation of the echo attenuation module 2 on a current frame and a previous frame as will be described here. -after in connection with Figures 4b and 4c.

  Figure 4b schematically shows the phenomenon of occurrence of pre- and / or post-echoes that occurs in two overlapping windows in a reverse transform decoder.

  In the aforementioned FIG. 4b: xm (n) denotes the decoded signal resulting from the inverse transform decoding for the current frame of rank m, current frame denoted Tc.

    Be denotes the current block of the current frame Tc comprising the samples between 0 and N / 2.

  Bs denotes Bs the following block of samples, i.e., the block of samples between the N / 2 and N samples of the current frame Tc.

  X, _, (n) denotes the inverse transformed decoded signal for the previous frame Tp comprising the samples of rank N / 2 to N / 2, the previous frame and the current frame being overlapped between the samples 0, N / 2 forming the previous block of the previous frame, block Bp overlapping with the current block Bc of the current frame Tc.

  FIG. 4b further shows a transition S1 as well as the noise level of the current frame of rank m.

  Figure 4c shows the value of the smoothed envelope signal ratio Rs (n) calculated as previously mentioned in the description.

  It can be seen from the observation of FIG. 4c and FIG. 4b that the pre-echo appears a half-window before the transition transition of the transition signal SI. The signals of the overlapping windows can then be cut or weakened before the rising transition in the case of the pre-echo. They are then cut off or weakened after the downward transition of the IF transition signal in the case of the post-echo.

  In the latter case, it is found that the echo persists half a window after the transition.

  The implementation of the device according to the invention according to the first embodiment of Figure 4a involves an adaptation of the calculation modules 10, 12 and 13 shown in Figure 3c in the manner below.

  The module 1 for calculating the echo decision, the echo origin position and the applicable attenuation value in the current frame is implemented so as to calculate a signal ratio of the echo. envelope smoothed from the auxiliary signal 4 (n) of the prediction decoder stage CELP together with the inverse transform decoded signal xTi (n) at the output of the stage MDCT transform decoding stage j and the ratio corresponding smoothed envelope.

  By way of nonlimiting example, it is indicated that the smoothed envelope signal is given by the relation: EnvTj (n) = [Max (x (n), EnvTi (n -1)] a + (1 a) [ 'cry (n) l.

  In the previous relation, it is indicated that the envelope signal of the inverse frequency transform decoded signal is computed recursively with respect to the value of the envelope signal for the preceding sample rank n-1, ie that is to say from the value EnvTj (n-1) by the choice of the maximum between this value and the absolute value of the decoded signal x,: j (n) by inverse transform of the current frame, by linear combination with the value aforementioned absolute with a designated experimental weighting factor a.

  In addition, as represented in FIG. 4d, the module 11 of the envelope ratio calculation makes it possible to calculate the smoothed envelope ratio according to the preceding relation from the envelope signals delivered by the module 10 delivering the signals of FIG. aforementioned envelope. The value of the envelope signal of the auxiliary signal delivered by the predictive decoder of rank i is calculated in the same manner as in the case of FIG. 3c.

  The value of the smoothed envelope signal ratio satisfies the relationship: EnvT (n) R15 (n) = n = 0, ..., KN -1. Env ,, (n) The aforesaid value is calculated for a frame of length KN where K denotes the total number of sub-blocks in a frame of length KN, N being the length of the subblock in number of samples on which is calculated a point of the envelope.

  As regards the calculation module specific for the attenuation value applicable in the current frame, it is indicated that the aforementioned calculation module is a software module which is adapted so as to allow the application of the attenuation value. given the existence of pre- or post-echoes as will be described below in connection with Figure 4e.

  The implementation of the calculation of the attenuation value, as described previously in the description, can be carried out either in the echo and echo origin decision decision module 1 or , if necessary, in the echo attenuation module 2 in a non-limiting manner.

  In general, with reference to FIG. 4e, it is indicated that for a decoded signal delivered by the inverse transform decoding module spectral coefficients formed by a current frame and a previous frame in recovery, as represented in FIG. 4b, the existence decision calculation module, the echo origin position module and the applicable attenuation value and, in particular, the module 13 of the latter may advantageously comprise as shown in FIG. a pre- or post-echo discrimination module in the preceding frame respectively in the current frame by comparison of the energy of the decoded samples of the preceding frame with the energy of the decoded samples of the current frame.

  For energies E _cur and E _ prev, signals in overlap xm j (n) of the current frame respectively xn, _1 j (n) of the previous frame the corresponding energies are calculated by the following relation: N! 2 1 E_cur = E x; 22 (n) n = 0 and N! 2 1 E _ prev = E x;, Z 1 (n). n = o If the energy of the current frame E_cur is greater than the energy of the previous frame E _ prev, then the existence of pre-echoes is detected when the value of the ratio of the envelope signals and, in particular, the value of the ratio of the smoothed envelope signals becomes greater than the threshold value S. In the opposite case, it is then in a post-echo situation being specified with reference to FIG. 4b that the transition SI intervening on the current frame of rank m the post-echo concerns in particular the sample block N to 3N / 2-1 of the following frame not shown in the drawing.

  A specific, particularly advantageous, nonlimiting procedure for calculating the attenuation values applicable to the current frame is explained below with reference to FIG. 4e.

  Following an initialization step 400 at the value of the sample rank n = 0 comparable to the step 202 of FIG. 3e, the aforementioned initialization step is followed by a step 401 of discriminating the presence of pre- and post-echo respectively. Step 401 can advantageously be performed as mentioned previously in the description, by comparing the energies of the current frame and the previous frame. The comparison deals with the values E _ cur and E _ prev.

  On discrimination of the existence of pre-echoes, the energy of the current frame being greater than the energy of the preceding frame, a test step 402 is called which consists in comparing the value of the ratio of the smoothed envelope signal R1; 5 (n) compared to the threshold value S. On negative response to the test 402, that is to say in the absence of a validly detected transition, a step 403 is called which maintains the value of the attenuations for the current frame and the previous frame, the attenuation value denoted gnij (n) for the previous frame and gnj (n) for the current frame at the value 1, no attenuation being introduced. Step 403 is followed by step 404 returning to step 402 to continue the test and incrementing the rank of sample n to the value n + 1.

  On the contrary, on a positive response to the test 402 a step 405 is called which consists in comparing the rank of the sample n with the value N / 2 end value of the current block B, of the current frame Tc.

  On a positive response to the test 405, a step 407 is called, which consists in assigning to the attenuation value for the previous frame g, n_lj (n) and in the value the attenuation values for the current frame gnj (n). the specific values C1 and C2, respectively. The values C1 and C2 are experimental specific values taking into account the windowing applied to the decoding.

  Step 407 is followed by step 408 of comparing the rank value of sample n with a maximum value N ,,, ax by comparison of superiority at step 408.

  On a negative response to the test 408, a step 409 is called which increments the value of the sample rank n = n + 1. The aforementioned step 409 is followed by a return to step 402 for further processing as long as the rank of the sample n is not greater than the value Nmax above.

  On the contrary, on a positive response to the test 408, a step 410 of end of calculation of the gain is called.

  On a negative response to the test 405, a step 406 is called which consists in keeping the attenuation value of the previous frame gn_4 (n) equal to the value 1, no attenuation being applied to the latter, and to apply to the current frame the attenuation value g, ni (n) taken for example at a value C3.

  Step 406 is then followed by a return to step 408 for a continuation of the process according to the test performed in step 408 previously described.

  On the discrimination of the presence of post-echoes in response to test 401, the energy of the current frame E_cur being lower than the energy of the preceding tram E_ prev, the process of calculating the attenuation value applied to the current frame respectively to the previous frame, is substantially identical to the described process relative to the detection of pre-echoes.

  Thus, on a positive response to the post-echo presence test 401, a test step 411 is called which substantially corresponds to the test step 402 previously described. On a negative response to the test 411 a step 412 and called which is to maintain the attenuation value of the current frame and the previous frame to the value 1, that is to say in the absence of attenuation, similar manner to step 403 previously described. Step 413 is also similar to incrementing the rank of sample n to the value n + 1 of step 404.

  On the contrary, on a positive response to the test 411, a step 414 is called to compare the sample rank n with the N / 2 value similar to the step 405 previously described. The test of step 414, however, is inverted with respect to the test of step 405. In fact, on negative response to test 414, there are post-echoes for the current frame and the previous frame. The step 416 called allows, in this case, to assign an attenuation value denoted gii * (n) = C4 respectively gnu (n) = C5. The values of C4 and C5 are specific experimental values.

  Step 416 is then followed by a step 417 of comparing the rank of the sample n by comparison of superiority with the maximum value Nmax at step 417. On a negative response to the test of step 417 a step 418 of incrementation n = n + 1 of the rank of the sample is called then followed by a return to the test of step 411 for continuation of the test and the aforementioned process.

  On the contrary, on positive test response 417 the end of the current frame has been reached, so an end of process step 419 is called.

  On a positive response to the test 414, a step 415 is called which consists of keeping the attenuation value of the current frame gmj (n) equal to the value 1, no attenuation being applied to the latter and to apply to the frame previous, the attenuation value gm_ij (n) taken for example at a value C6.

  Step 415 is then followed by a return to step 417 for a continuation of the process according to the test performed in step 417 previously described.

  For a more detailed description of the choice of the value of the attenuation coefficient C1, C2, C3, C4, C5 and C6, it is useful to refer to the French patent application FR 05 04572 filed in the name of the applicant on 4 May 2005.

  Finally, with reference to FIG. 4f, the attenuation module 2 may advantageously be constituted, when the latter is implemented upstream of the overlap addition module, by direct application of the attenuation values, ie say the values g, 1 (n) and g (n) by means of multipliers 21 and 22 respectively directly applied to the decoded signal by inverse transform, that is to say to the signals xn, 1 (n) and respectively x11 (n) as shown in Figure 4f.

  The multipliers 21 and 22 respectively deliver a cut signal in which the echoes have been attenuated, each corresponding signal delivered being denoted xrmD j (n) respectively xn, _, D j (n) which are directly delivered to the overlap adding module to reconstitute the reconstructed signal x; d (n). In Fig. 4a, the dashed arrow from the predictive coding portion to the decoding portion of the spectral coefficients represents the possibility of reconstructing the spectral coefficients using partially the predictive decoder output. By way of nonlimiting example, a solution consists in performing a time-frequency transformation of the output signal of the predictive coder and then in subtracting decoded spectral coefficients, the spectral components of the predictive signal that it is desired to eliminate from such a fate that the signal reconstituted time XS (n) can be obtained by adding the output signal of the predictive encoder and the output signal of the frequency-time transformation after attenuation. Indeed, this procedure is effective because the received transform coefficients represent the signal and they can advantageously replace the spectral coefficients of the predictive signal in certain frequency areas. The pseudo code for performing the operation "decoding spectral coefficients" is identical to that described for Figure 3b.

  A second alternative embodiment of the device, object of the present invention, of simplified design compared to the embodiment of the device object of the invention shown in Figure 4a, will now be described in connection with Figure 5a and the Figure 5b.

  FIG. 5a considers a reverse-transform decoder of rank j in which the auxiliary addition / overlap module is represented in FIG. 4a is deleted. Under these conditions, the inverse MDTC decoding module directly delivers the decoded signals xn, j (n) for the current frame and x13 (n) of the previous frame to the attenuation module of the echo 2, such as 4f, for example, which delivers the aforementioned signals itself to the addition / overlap module.

  Under these conditions, the module 1 for calculating the existence of the original position of the echo and the attenuation value delivering the signal DPG also receives a current frame and a previous frame successively of transform-decoded spectral coefficients. inverse in the absence of auxiliary addition / recovery.

  Thus, the calculation module 1 is adapted to include, as shown in FIG. 5b, the temporal envelope calculation module 1 o, the envelope ratio calculation module 11 and the module 12. calculation of pre-and post-echo discrimination and calculation of the envelope ratio under the conditions below.

  The temporal envelope calculation module 10 receives the auxiliary signal decoded by the signal predictive decoder 4 (n), the decoded signal representative of the current frame xm j (n) and the decoded signal representative of the preceding frame, xm_, j (not) . The module 1 o then makes it possible to jointly examine the previous frame xm_1 j (n) and the aforementioned current frame xm] (n) which is in overlap, as represented in FIG. 4b. It delivers the envelope signals Envp, (n), Envmj (n) for the current frame respectively, Envm_lj (n) to the calculation module of the envelope ratio 11.

  The module 11 for calculating the envelope ratio then makes it possible to calculate values of first envelope ratio Rm_1 j (n) between the decoded signal representative of the previous frame xrn_1 (n) and the decoded auxiliary signal xp, (n) on the one hand, and second envelope ratio Rn, (n) between the decoded signal of the current frame x, j (n) and the auxiliary decoded signal delivered by the predictive decoder xp; (n) on the other hand.

  As in the case of FIG. 3e for example, the attenuation values applicable to the current frame and to the previous frame, values gm i (n) respectively g_11 (n) are then deduced from the envelope ratios R i ,, i (n) and R ,,; _, j (n) respectively. In particular, the attenuation values can be constituted by the inverse of the values of the aforementioned envelope ratios.

  From the aforementioned envelope ratio values, the first time envelope ratio of the previous frame to the time envelope of the auxiliary signal and the second time envelope ratio of the current frame to the time envelope of the auxiliary signal, a pre-echo and / or post-echo presence discrimination calculation module from the abovementioned first and second envelope ratio ratios then makes it possible to execute a process similar to that described with reference to FIG. pre-purposely post-echoes.

  By way of nonlimiting example, if the second envelope ratio R1 (n) of the current frame to the time envelope of the auxiliary signal is greater than a threshold value and the value of the first time envelope ratio R , _11 (n) is weak, the presence of pre-echoes is then detected. If the first and second envelope ratios Rn, j (n) and i (n) are both greater than a threshold value, then the transition signal SI corresponding to an attack transition or a decay transition is located in the corresponding current frame.

  If the first time envelope ratio t "_1 j (n) is greater than a threshold value and the second ratio R i (n) has a low value, then there are post-echoes.

  The reduction of pre- and post-echoes can be performed in a manner similar to that carried out and described in connection with FIG.

  However, in the case of the implementation of FIG. 5a of the device that is the subject of the invention, the smoothed envelope ratio is replaced by the value Rc j (n) taking into account the discrimination of the pre- and post-echoes as previously described.

  The attenuation values are then directed to the multipliers 21 and 22 of Figure 4f to perform the reduction of echoes.

  The slicing in the window where the attack or decay SI transition occurs can, however, be refined as follows: In the frame where the attack or decay transition occurs, it is possible to calculate the ratio envelope between the auxiliary signal xP; (n) and the signal sum of the decoded signal representative of the current frame and the preceding frame, the aforementioned sum signal satisfying the relation: xsTJ (n) = x ,,, (n) + x, 1 (n) É Le The above procedure is justified by the fact that the envelope of this sum of signals is compared with that of the auxiliary signal x P1 (n). In the frame where the transition S1 as shown in FIG. 4c occurs, before the aforementioned transition, the folding terms cancel each other out because they are in phase opposition in accordance with the relation: xn, (n ) = h (n) x (n) h (N / 2 1 n) x (N / 2 1 n) n = 0, ..., N / 2 1 and xn (n) = h (n) x ( n) + h (3N 1 n) x (3N 1 n) n = N / 2, ..., N 1 When the transition occurs, the non-folded terms are then very large in front of the folded terms and the echo can thus be detected.

  The invention also covers a message indicating presence or absence of echoes in a current frame of digital audio signal message M (Ec = [0,1]) as described in particular in Figure 2c. The message may be embedded in a current frequency-coded digital audio signal frame and is a false alarm signaling message. It comprises at least one field comprising a bit at the value 0 or 1 determined to indicate an a priori echo absence on a current frame of digital audio signal coded respectively to a complemented value to indicate an a priori presence of echo on a frame current of coded digital audio signal.

  In Fig. 5a, the dashed arrow from the predictive coding portion to the decoding portion of the spectral coefficients represents the possibility of reconstructing the spectral coefficients using partially the predictive decoder output. By way of non-limiting example, one solution consists in performing a time-frequency transformation of the output signal of the predictive coder and then in subtracting decoded spectral coefficients, the spectral components of the predictive signal that it is desired to eliminate from such a fate as the reconstituted temporal signal. XS (n) can be obtained by adding the signal at the output of the predictive coder and the signal at the output of the transformation frequency and time after attenuation. Indeed, this procedure is effective because the received transform coefficients represent the signal and they can advantageously replace the spectral coefficients of the predictive signal in certain frequency areas. The pseudo code for performing the "decoding spectral coefficients" operation is identical to that described for FIG. 3b.

  Finally, as represented in FIGS. 6a and 6b, the device that is the subject of the invention further comprises, at the level of the coder, a module for calculating an echo signaling message M (Ec) in the frame common. The signaling message is transmitted together with the current frame.

  In particular, for a multilayer encoder comprising at least one predictive encoder and a rank transformation encoder j, the calculation module of a signaling message comprises at least, as represented in FIG. 6a, a module for calculating a signal. difference between a reconstituted auxiliary decoded signal noted xP; (n) this signal, as represented in FIG. 6a above, being obtained by predictive coding, then predictive decoding of the digital audio signal before encoding x (n) and the digital audio signal before coding delayed by a duration equal to the algorithmic delay of the predictive coding-decoding part (anticipation or look-ahead in English and delay of the under- and oversampling filters) via a delay circuit ZT. The delayed signal x (n T) is delivered to a subtracter circuit which outputs the difference signal x D, e (n). In addition, the module for calculating the signaling message comprises at least, as represented in FIG. 6b, a module for calculating the time envelope values of the reconstituted auxiliary decoded signal, the signal xP1 (n) and the temporal envelope. the difference signal xo.j (n). It is understood, in particular, that the module inserted in the encoder substantially corresponds to the module lo described above in the description in connection with Figure 3b for example.

  The module 1 o inserted in the encoder delivers the envelope of corresponding signals denoted Envp, (k) corresponding to the time envelope signal of the reconstituted auxiliary decoded signal and the time envelope signal of the difference signal denoted Envo; f j (k). The module for calculating the signaling message further comprises a module 1 I for calculating the envelope ratio comparable to that previously described in the description in conjunction with FIG. 3b.

  Finally, the module 13 makes it possible to calculate the value of the ec ec ec ecence presence variable Ec, this module being comparable to that described above in the description for the echo presence detection from the comparison of the value of the envelope envelope ratio. the threshold value S in connection with FIG. 3b, 3c and 3d, for example.

  In the configuration according to FIG. 3a, it is possible that in certain frequency zones the predictive coder does not lead to a sufficiently precise coding. In this case, it is possible to directly code these areas of the spectrum by the transform coding stages. At the decoder, when the transformation enhancement layers are received, they replace the corresponding areas of the spectral coefficients resulting from the time-frequency transformation of the output signal of the predictive decoder. In this case, the reconstituted signal XS (n) is obtained by means of a frequency-time transformation. The echo attenuation device processes the echoes generated by this transformation. In relation to the preceding cases, several variants have been considered, with reference to FIGS. 7 to 10, the same references designate the same elements as in the preceding figures, which for this reason will not be described in detail. In FIGS. 7 to 10, the spectral coefficients received at the decoder may represent in certain frequency zones (typically in the high frequencies) the original signal to be coded and not the difference signal as is typically the case in the low frequencies. Indeed, in high frequencies, some areas are not modeled with sufficient accuracy by the predictive coder, in these areas it is more interesting to encode by the transform encoder the original signal instead of the difference signal. At the decoder, when these transform coefficients are received, they replace the coefficients from the predictive decoder.

  Figure 7 shows a variant of Figure 3b. In this figure, the signal coming from the predictive decoder is passed in the frequency domain, here the Transformed Time / Frequency operation MDCT delivering the Coeffpi coefficients (k), then combined with the CoeffTi (k) spectral coefficients and finally passed in the time domain. by inverse transform and addition / recovery operations. The echo attenuation is performed after the addition / overlap.

  Figure 8 shows a variant of Figure 4a. In this figure the signal from the predictive decoder is passed in the frequency domain, here the MDCT operation, and then combined with the spectral coefficients. A Frequency / Time transformation is performed. The output is used on the one hand to supply the addition / recovery and then the echo decision device; on the other hand to supply the echo attenuation device whose output produces the decoded signal after recovery. The echo attenuation is performed just before the addition / overlap.

  Figure 9 shows a variant of Figure 5a. In this figure the signal from the predictive decoder is passed in the frequency domain, here the MDCT operation, and then combined with the spectral coefficients. A Frequency / Time transformation is performed. The output is used on the one hand to supply the echo decision device on the other hand to supply the echo attenuation device whose output produces the decoded signal after recovery. The attenuation of the echoes is carried out just before the addition / overlap, with the decision taken on the output signals of the decoding stage of the spectral coefficients and inverse MDCT.

  The pseudo code for performing the operation "combination of spectral coefficients" of Figures 7, 8 and 9 is given below.

  Nickname Code for k = 0, ..., Kb-1 CoeffRi (k) = Coeffp; (k) + CoeffTT (k) end for k = Kb, ..., Kf-1 if (flag (k) == available) - 40 -CoeffRR (k) = CoeffTi (k) else CoeffRR (k) = Coeffp (k) endif end In the embodiment represented in FIGS. 7, 8 and 9, it is noted that the somerator So of FIGS. 3b, 4a and 5a are no longer necessary, the corresponding processing can be performed on the spectral coefficients.

  Fig. 10 shows a variant of Fig. 3b and 7 where echo detection and attenuation are separately on a plurality of sub-bands of the transform and decoded signal. It may be noted that the subband filtering is carried out in the present case by an inverse MDCT and an addition / overlap respectively from the spectral coefficients of the low band Coeffeb (k) and the high band CoeffRJBh (k). This subband filtering can also be carried out in a non-limiting and alternative manner directly on the time signal resulting from a frequency / time transposition operation and addition recovery of the recombined spectral coefficients CoeffRj (k). Here, the output of the predictive stages has been separated into two bands, low band and high band, in order to feed the decision and echo attenuation device respectively of the low band and the high band of the decoded transform signal. . In FIG. 10, the device that is the subject of the invention thus comprises a first low-band echo attenuation channel and a sub-band echo attenuation channel. The constituent modules of the aforementioned channels are similar but dedicated to the respective frequencies of each sub-band.

  In addition, according to the embodiment of FIG. 10, the number of attenuation channel processing subbands can be arbitrary.

  The pseudo code for carrying out the "combination of the spectral coefficients" operation of FIG. 10 is given below.

  Pseudo Code for k = 0, ..., Kb-1 CoeffRiBb (k) = CoeffTi (k) Coeffeh (k) = 0 -41 - end for k = Kb, ..., Kf-1 if (flag (k ) == dispo) Coeffeh (k) = CoeffTT (k) else Coeffeh (k) = Coeffp (k) endif Coeffeb (k) = 0 end Figure 11 represents a variant of Figure 10 where the predictive signal summed by the summator So is not filtered contrary to the configuration of Figure 10 and where the predictive signal is filtered into a plurality of sub-bands by digital filters to power the echo detection device. Thus, the predictive signal is not filtered but directly added in the So adder to the signals derived from the transform decoding after filtering in the frequency domain. In counter part the pseudo code of the operation "combination of the spectral coefficients" is modified as follows: Pseudo Code for k = 0, ..., Kb-1 Coeffeb (k) = CoeffTi (k) CoeffRJBh (k) = 0 end for k = Kb, ..., Kf-1 if (flag (k) == dispo) CoeffRiBh (k) = CoeffTJ (k) - Coeffp; (k) else CoeffRiBh (k) = 0 endif Coeffeb ( k) = 0 end An advantage of this way of proceeding is that the signals summed by So are in phase and that the low-pass and high-pass filters of the echo detectors can be carried out indifferently by impulse response filters. - 42 - linear phase or infinite impulse response filters possibly with minimal phase shift.

  In the embodiment described above, the predictive signal is directly added to the signals of the sub-bands resulting from the transform decoding and the combination of the spectral coefficients subtracts the spectral components of the predictive signal that it is desired to replace by those signal from transform decoding.

  The invention also encompasses a computer program product recorded on a storage medium and executable by a computer, for echo discrimination of a multi-layered hierarchical coded digital audio signal from a generator-generated transform coding. echoes, and non-echo-generating predictive coding.

  It is remarkable that the program product, object of the invention, when executed by a computer allows the implementation of the method as described above in the description in conjunction with Figures 2a, 2b and 2c.

  The computer program product, which is the subject of the invention, is furthermore remarkable in that it is implanted in the echo presence calculation module, the echo origin position module and the value value module. echo attenuation applicable to the current frame, as described and referenced 1 relative to the device object of the invention previously described in connection with FIGS. 3b, 3c to 3f, 4a to 4e and 5a, 5b and / or 7 to 10 The computer program product, object of the invention, is integrated in a multilayer hierarchical decoder, a digital audio signal, as described previously in the description in conjunction with FIGS. 6a and 6b. The echo existence discrimination module makes it possible to generate a difference signal the signal xD i (n) from a predictive decoding coding of this digital signal and a presence variable respectively of absence of echo in the form of a false alarm message transmitted with the current frame, as previously described in the description.

- 43 -

Claims (23)

  A method of discriminating the echoes of a digital audio signal generated by multilayer hierarchical coding from a transform coding, an echo generator, and a predictive, non-echo generating coding, characterized in that said method is at least decoding, for each current frame of said digital audio signal, to: compare, in real time, a variable resulting from a characteristic of the temporal envelope of the signal resulting from an echo generator decoding to the corresponding characteristic of the signal from non-echo decoding at a threshold value; and, depending on the result of the comparison to the threshold value, conclude whether or not there is an echo from transform coding in said current frame 2. Method according to claim 1, characterized in that the original position of said echo in said current frame is given by the position in said current frame of said variable substantially equal to said threshold value.   3. Method according to one of claims 1 or 2, characterized in that it further comprises, coding, for each transmitted frame, to: detect the existence of echo in the current frame; and transmitting together said decoding said current frame and an echo-existence signaling message, whereby the echo discrimination is decoded conditionally on the existence or non-existence, a priori, of echo to coding.   4. Method according to claim 3, characterized in that said existence and echo position signaling message is a false alarm message of presence of echoes in said current frame.   A device for discriminating and attenuating the echoes of a digital audio signal generated by a multilayer hierarchical coder, said encoder comprising a first plurality of predictive, non-echo-generating decoders, and a second plurality of transform decoders capable of revealing echoes, each predictive encoder having at least one output outputting a decoded predictive digital audio signal which may be either said decoded digital audio signal or said decoded and filtered digital audio signal or a signal representative of the energy of said decoded digital audio signal, each transform decoder delivering, after decoding by inverse transforming and overlaying, a decoded signal sequence of current frames, characterized in that said device comprises at least at the level of a decoder: a means for calculating the existence and the position of Echo origin in the current frame receiving, on the one hand, the decoded predictive signal delivered by the output of a predictive decoder of rank i of this plurality of predictive decoders, and, on the other hand, the decoded signal delivered by the output of a decoder by transform inverse of rank j of this plurality of decoders by inverse transform and delivering a presence signal, echo position and echo attenuation value applicable in said current frame; and, echo attenuation means receiving said decoded current frame succession signal from said row-rank inverse transform decoder and said presence, echo origin position and attenuation value signal. echo applicable in said current frame and delivering a decoded signal current frame succession in which the echo is attenuated from the original position of the echo.   6. Device according to claim 5, characterized in that said decoded signal, in successive frames with an attenuated echo, is followed by inverse perceptual filtering.   7. Device according to claim 5, characterized in that said means for calculating the existence and the original position of the echo in the current frame comprises at least: means for calculating the time envelope of said signal predictive decoded output from the output of said rank i predictive decoder and the time envelope of the decoded signal output from the decoder output by rank transform j; Means for calculating the ratio of said time envelope values; means for calculating the existence decision, the echo origin position and the attenuation value applicable to said echoes by comparing the value of said ratio of said time envelope values with a threshold value.   8. Device according to claim 7, characterized in that said attenuation value is either the value 1, in the absence of echo, or the value of the inverse of the ratio of said time envelope values.   9. Device according to one of claims 5 to 8, characterized in that it further comprises at said decoder means for receiving and processing an echo signaling message transmitted jointly with each current frame of the encoder to said decoder.   10. Device according to claim 9, characterized in that, for a message message of the false alarm message type of presence of echoes in said current frame, said device comprises associated with said position-existence decision calculating means d origin of echo and attenuation value applicable to said echoes: means for reading said false alarm message; and, on reading a false alarm message at the true value, the information delivered by said message corresponding to an a priori echo absence on said current frame, means for inhibiting said decision calculating means of existence, echo origin position and applicable attenuation value; otherwise, the information delivered by said message corresponding to an a priori echo presence on the current frame, activation means of said echo origin position value decision calculation means and applicable attenuation.   11. Device according to one of claims 5 to 10, characterized in that for a reverse-transform decoder of rank j comprising an inverse frequency decoding module of the spectral coefficients connected in cascade with an addition / recovery module, said decoder comprises: - said echo attenuation means interposed between said inverse decoding module spectral coefficients and said addition / recovery module; an auxiliary addition / recovery module receiving two successive frames of decoded spectral coefficients delivered by said inverse decoding module and delivering an auxiliary decoded signal in which the echoes are not attenuated to said means of calculating the existence, the position of d origin of the echo in the current frame, said means for calculating the existence, the original position of the echo and the attenuation value in the current frame delivering said position presence signal of echo and attenuation value in the current frame to said echo attenuation means.   12. Device according to claim 11, characterized in that for a decoded signal delivered by said inverse frequency decoding module spectral coefficients formed by a current frame and a previous frame overlay, said position existence decision calculating means echo origin and the applicable attenuation value further comprises means for discriminating pre-respectively post-echoes in the previous frame respectively in the current frame, by comparing the energy of the decoded samples of the frame preceding the energy of the decoded samples of the current frame, said current frame being the seat of pre-echoes for an energy of the decoded samples of the current frame greater than the energy of the decoded samples of the previous frame, when the ratio values of said envelope signals are greater than said threshold value, said current frame being e post-echo seat, otherwise.   13. Device according to claim 11, characterized in that for a row-inverse transform decoder in which said auxiliary addition / overlap module is suppressed, said means for calculating the existence, the position of origin of the echo and the applicable attenuation value in the current frame receives a current frame and a previous frame of inverse transform decoded spectral coefficients, in the absence of addition, said calculating means comprises: means calculating the temporal envelope of the preceding frame, the current frame and the auxiliary signal delivered by the second output of the predictive decoder of rank i; means for calculating a first time envelope ratio of the preceding frame to the time envelope of the auxiliary signal and a second time envelope ratio of the current frame to the time envelope of the auxiliary signal, means of discriminating the presence of pre-echo or post-echo respectively from said ratios and of calculating attenuation values, taking into account the presence of post-echo preprespectively and the value of said first and second ratios.   14. Device according to one of claims 9 to 13, characterized in that it further comprises, at the coder, means for calculating an echo signaling message, said signaling message. being transmitted together with the current frame.   15. Device according to claim 14, characterized in that, for a multilayer encoder comprising at least one predictive coder of rank i and a coder by rank transform j, said means for calculating a signaling message comprise at least: means for calculating a difference signal between a reconstructed auxiliary decoded signal, obtained by coding-decoding of the digital audio signal by means of said rank i predictive coder and of a predictive decoder, and the delayed digital audio signal of a duration equal to the delay algorithmic of the predictive coding-decoding part; means for calculating the values of the temporal envelope of said reconstructed auxiliary decoded signal and of the temporal envelope of said difference signal; means for calculating the ratio of said envelope values.   - 48 - 16. Device according to one of claims 5 to 15 above, characterized in that said device further comprises, at the decoder, means for reconstructing the spectral coefficients of the decoder by inverse transform of rank j, from a at least part of the output signal delivered by said rank i predictive decoder.   17. Device according to claim 16, characterized in that said means for reconstituting the spectral coefficients comprise at least: time-frequency transformation means of the output signal of the predictive decoder delivering the spectral components of the predictive signal; means for combining the spectral components of the predictive signal with the decoded spectral coefficients.     means for separating said decoded spectral coefficients into a plurality of subbands by frequency / time and addition / overlap transposition.     means for separating the predictive signal into a plurality of sub-bands 18. Device according to claim 17, characterized in that said means for combining the spectral components of the predictive signal with the decoded spectral coefficients deliver at least: first combined spectral coefficients in a determined frequency band and a temporal signal corresponding to said band of determined frequency; second spectral coefficients combined in at least one other determined frequency band and the time signals corresponding to said determined frequency bands; said device having an echo detection and attenuation path in said frequency band respectively at least one other echo detection and attenuation path in each other frequency band.   19. Device according to one of the preceding claims 16, 17, 18, characterized in that the predictive signal is directly added to the signals of the sub-bands resulting from the transform decoding and in that the combination of the spectral coefficients realizes a subtraction of the spectral components of the predictive signal that it is desired to replace by those of the signal resulting from the transform decoding.   20. The presence or absence of echo signaling message in a current frequency-transform coded digital audio signal frame, characterized in that said message constitutes a false alarm signaling message, said message comprising at least one field comprising a bit-value bit determined to indicate an a priori echo absence on a current encoded complement-coded digital audio signal frame respectively to indicate an echo prior presence on a coded digital audio signal current frame.   21. Computer program product recorded on a computer-executable storage medium for echo discrimination of a multi-layer hierarchical encoded digital audio signal from an echo generator transform coding, and non-echo-generating predictive coding, characterized in that said program product when executed by a computer allows the implementation of the method according to one of claims 1 to 4.   22. Computer program product recorded on a storage medium, executable by a computer, according to claim 21, characterized in that it is implanted in a module for calculating the presence of echo, of the original position of the computer. the echo and echo attenuation value applicable to the current frame, integrated into a multilayer hierarchical decoder, said module receiving, on the one hand, a decoded signal representative of the energy of the signal delivered by a predictive decoder of rank i, and, on the other hand, at least one decoded signal delivered by a reverse transform decoder of rank j.   23. Computer program product recorded on a storage medium, executable by a computer, according to claim 21, characterized in that it is implanted in an echo-existence discrimination module integrated into a multilayer hierarchical coder. , a digital audio signal, said module - 50 - for generating a difference signal from a predictive decoding coding of this digital signal and a presence variable respectively absence of echo as a message of false alarm transmitted with the current frame.