EP3192073B1 - Discrimination and attenuation of pre-echoes in a digital audio signal - Google Patents

Description

The invention relates to a method and a device for discriminating and processing pre-echo attenuation when decoding a digital audio signal.

For the transmission of digital audio signals over telecommunications networks, whether for example fixed or mobile networks, or for the storage of signals, compression processes (or source coding) using coding systems which are generally of the linear coding type by linear prediction or by transform frequency coding.

The method and the device, which are the subject of the invention, thus have as their field of application the compression of sound signals, in particular frequency-coded digital audio signals.

The figure 1 represents for illustrative purposes, a block diagram of the coding and decoding of a digital audio signal by transform including an addition / overlap synthesis analysis according to the prior art.

Certain musical sequences, such as percussion and certain segments of speech like the plosives (/ k /, / t /, ...), are characterized by extremely sudden attacks which result in very fast transitions and a very strong variation signal dynamics in a few samples. An example of a transition is given to the figure 1 from sample 410.

For the coding / decoding process, the input signal is cut into blocks of samples of length L whose boundaries are represented on the figure 1 by dotted vertical lines. The input signal is denoted x ( n ), where n is the index of the sample. Cutting into successive blocks (or frames) leads to defining the blocks X _N ( n ) = [ x (NL) ... x (N.L + L-1)] = [ x _N (0) ... x _N (L-1)], where N is the index of the block (or the frame), L is the length of the frame. To the figure 1 we have L = 160 samples. In the case of the modulated modified cosine transform MDCT (for "Modified Discrete Cosine Transform" in English), two blocks X _N (n) and X _{N + 1} (n) are analyzed together to give a block of transformed coefficients associated with the frame of index N and the analysis window is sinusoidal.

The division in blocks, also called frames, operated by the transform coding is totally independent of the sound signal and the transitions can therefore appear at any point in the analysis window. But after transform decoding, the reconstructed signal is tainted by "noise" (or distortion) generated by the quantization (Q) -quantization inverse (Q ^-1 ) operation. This coding noise is temporally distributed relatively uniform throughout the temporal support of the transformed block, that is to say over the entire length of the window length 2L samples (with overlap of L samples). The energy of the coding noise is generally proportional to the energy of the block and is a function of the coding / decoding rate.

For a block with an attack (such as block 320-480 of the figure 1 ) the signal energy is high, so the noise is also high.

In transform coding, the level of the coding noise is typically lower than that of the signal for the high energy segments that immediately follow the transition, but the level is higher than that of the signal for the lower energy segments, especially on the part preceding the transition (samples 160 - 410 of the figure 1 ). For the aforementioned part, the signal-to-noise ratio is negative and the resulting degradation can appear very troublesome to listen. Pre-echo is the coding noise prior to the transition and post-echo the noise after the transition.

We can observe on the figure 1 that the pre-echo affects the frame preceding the transition as well as the frame where the transition occurs.

Psychoacoustic experiments have shown that the human ear performs a rather limited temporal pre-masking of sounds, of the order of a few milliseconds. The noise preceding the attack, 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 inconvenience 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 well known that for stationary signals plus the length of the transform increases, the higher the coding gain. At a fixed sampling rate and fixed rate, if we increase the number of points of the window (hence the length of the transform) we will have more bits per frame to code the frequency lines found useful by the psychoacoustic model, d where the advantage of using blocks of great length. MPEG AAC (Advanced Audio Coding) coding, for example, uses a long window that contains a fixed number of samples, 2048, or a duration of 64 ms if the sampling frequency is 32 kHz; the problem of pre-echoes is managed by allowing to switch from these long windows to 8 short windows through intermediate windows (called transition), which requires some delay in coding to detect the presence of a transition and adapt the windows. The length of these short windows is therefore 256 samples (8 ms at 32 kHz). At low speed you can always have an audible pre-echo of a few ms. The switching of the windows makes it possible to attenuate the pre-echo but not to suppress it. Transform encoders used for conversational applications, such as ITU-T G.722.1, G.722.1C or G.719, often use a 20 ms frame length and a 40 ms window at 16, 32 or 48 kHz (respectively). It should be noted that the ITU-T G.719 encoder incorporates a window switch mechanism with transient detection, however the pre-echo is not completely reduced at low bit rate (typically at 32 kbit / s).

In order to reduce the aforementioned annoying effect of the pre-echo phenomenon, various solutions have been proposed at the encoder and / or the decoder.

Window switching has been mentioned previously; it requires transmitting auxiliary information to identify the type of windows used in the current frame. Another solution is to apply adaptive filtering. In the area preceding the attack, the reconstructed signal is seen as the sum of the original signal and the quantization noise.

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

The implementation of such a filtering requires the knowledge of parameters some of which, like the prediction coefficients and the variance of the signal corrupted by the pre-echo, 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. This requires additional information to be transmitted, which constrains the relative budget allocated to the transform coding. 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. However, it requires transmitting the additional parameters to the decoder.

Unlike previous solutions, different pre-echo reduction techniques without specific transmission of information have been proposed. For example, a review of pre-echo reduction in the context of hierarchical coding is presented in the article B. Kövesi, S. Ragot, M. Gartner, H. Taddei, "Pre-echo reduction in the ITU-T G.729.1 embedded coder," EUSIPCO, Lausanne, Switzerland, August 2008 .

A typical example of pre-echo attenuation method without auxiliary information is described in the French patent application FR 08 56248 . In this example, sub-block attenuation factors are determined in the low energy sub-blocks preceding a sub-block in which a transition or attack has been detected.

où f est une fonction décroissante à valeurs entre 0 et 1 et k est le numéro du sous-bloc. D'autres définitions du facteur g(k) sont possibles, par exemple en fonction de l'énergie En(k) dans le sous-bloc courant et de l'énergie En(k-1) dans le sous-bloc précédent.The attenuation factor g ( k ) in the k-th sub-block is calculated for example according to the ratio R ( k ) between the energy of the sub-block of higher energy and the energy of the k-th under -block in question: $$\mathrm{boy\; Wut}\left(k\right)=f\left(R\left(k\right)\right)$$

where f is a decreasing function with values between 0 and 1 and k is the number of the sub-block. Other definitions of the factor g ( k ) are possible, for example as a function of the energy En ( k ) in the current sub-block and the energy En ( k -1) in the preceding sub-block.

If the energy of the sub-blocks varies little with respect to the maximum energy in the sub-blocks considered at the current frame, no attenuation is then necessary; the factor g ( k ) is set to an attenuation-inhibiting attenuation value, i.e. 1. Otherwise, the attenuation factor is between 0 and 1.

In most cases, especially when the pre-echo is troublesome, the frame that precedes the pre-echo frame has a homogeneous energy that corresponds to the energy of a low energy segment (typically a background noise). According to experience it is not useful or even desirable that after pre-echo attenuation processing the signal energy becomes lower than the average energy (sub-block) of the signal preceding the treatment zone - typically that of the previous frame, noted In , or that of the second half of the previous frame, noted In .

où l'énergie moyenne du segment précédent est approximée par la valeur max ( En, En ').For the subscript subscript k to be processed, the limit value, denoted lim _g ( k ) , of the attenuation factor can be calculated in order to obtain exactly the same energy as the average energy per sub-block of the preceding segment. the sub-block to be processed. This value is of course limited to a maximum of 1 since we are interested here in the attenuation values. More precisely, we define here: $${\lim }_{\mathrm{boy\; Wut}}\left(k\right)=\min \left(\sqrt{\frac{\max \left(\widetilde{\mathit{In}},\widetilde{\mathit{In}}\text{'}\right)}{\mathit{In}\left(k\right)}}\mathrm{,\; 1}\right)$$

where the average energy of the previous segment is approximated by the max value ( In , In ').

The value lim _g ( k ) thus obtained serves as a lower limit in the final calculation of the attenuation factor of the sub-block, it is therefore used as follows: $$\mathrm{boy\; Wut}\left(k\right)=\max \left(\mathrm{boy\; Wut}\left(k\right),{\lim }_{\mathrm{boy\; Wut}}\left(k\right)\right)$$

The attenuation factors (or gains) g ( k ) determined by sub-blocks can then be smoothed by a sample-by-sample applied smoothing function to avoid abrupt changes in the attenuation factor at the block boundaries.

où L' représente la longueur d'un sous-bloc.
La fonction est ensuite lissée suivant l'équation suivante: $$g_{\mathit{pre}}\left(n\right):=\alpha g_{\mathit{pre}}\left(n-1\right)+\left(1-\alpha \right){\mathrm{g}}_{\mathit{pre}}\left(n\right),n=\mathrm{0,}\cdots ,L-1$$

avec la convention que g_pre (-1) est le dernier facteur d'atténuation obtenu pour le dernier échantillon du sous-bloc précédent, α est le coefficient de lissage, typiquement α=0.85.For example, we can first define the gain per sample as a piecewise constant function: $${\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}\right)=\mathrm{boy\; Wut}\left(k\right),\mathrm{}\mathrm{not}=\mathit{kL}\text{'},\cdots ,\left(k+1\right)\mathrm{The}\text{'}-1$$

where L 'represents the length of a sub-block.
The function is then smoothed according to the following equation: $${\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}\right):=\alpha {\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}-1\right)+\left(1-\alpha \right){\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}\right),\mathrm{not}=0\cdots ,\mathrm{The}-1$$

with the convention that g _pre (-1) is the last attenuation factor obtained for the last sample of the previous sub-block, α is the smoothing coefficient, typically α = 0.85.

où g_pre'(n) est l'atténuation non lissée et g_pre (n) est l'atténuation lissée, g_pre'(n) avec n=-(u-1),····,-1 sont les derniers u-1 facteurs d'atténuation obtenus pour les derniers échantillons du sous-bloc précédent. On peut par exemple prendre u = 5.Other smoothing functions are also possible, for example the linear cross fade on u samples: $${\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}\right)=\frac{1}{u}{\displaystyle \underset{i=0}{\overset{u-1}{\Sigma }}{\mathrm{boy\; Wut}}_{\mathit{pre}}\text{'}}\left(\mathrm{not}-i\right),\mathrm{}\mathrm{not}=0\cdots ,\mathrm{The}-1$$

where g _pre ' ( n ) is the unsmoothed attenuation and g _pre ( n ) is the attenuation smoothed, g _pre ' ( n ) with n = - ( u -1), ····, -1 are the last u-1 attenuation factors obtained for the last samples of the previous sub-block. We can for example take u = 5.

où x_rec,g (n) est le signal décodé et post-traité par la réduction de pré-écho.Once the factors g _pre ( n ) thus calculated, the attenuation of pre-echoes is made on the reconstructed signal in the current frame, x _rec ( n ) , by multiplying each sample by the corresponding factor: $$x_{\mathit{rec},\mathrm{boy\; Wut}}\left(\mathrm{not}\right)={\mathrm{boy\; Wut}}_{\mathit{pre}}\left(\mathrm{not}\right)x_{\mathit{rec}}\left(\mathrm{not}\right),\mathrm{}\mathrm{not}=0\cdots ,\mathrm{The}-1$$

where x _{rec, g} ( n ) is the signal decoded and post-processed by the pre-echo reduction.

The figures 2 and 3 illustrate the implementation of the attenuation method as described in the patent application of the state of the art, cited above, and summarized above.

In these examples the signal is sampled at 32 kHz, the length of the frame is L = 640 samples and each frame is divided into 8 sub-blocks of K = 80 samples.

In part a) of the figure 2 a frame of an original signal sampled at 32 kHz is shown. An attack (or transition) in the signal is located in the sub-block beginning at the index 320. This signal has been coded by a low rate (24 kbit / s) MDCT type transform coder.

In part (b) of the figure 2 , the result of the decoding without pre-echo processing is illustrated. The pre-echo can be observed from sample 160, in the preceding sub-blocks the one containing the attack.

Part c) shows the evolution of the pre-echo attenuation factor (solid line) obtained by the method described in the aforementioned prior art patent application. The dotted line represents the factor before smoothing. Note here that the position of the attack is estimated around the sample 380 (in the block delimited by the samples 320 and 400).

Part d) illustrates the result of the decoding after application of pre-echo processing (multiplication of signal b) with signal c)). We see that the pre-echo has been attenuated. The figure 2 also shows that the smoothed factor does not go back to 1 at the moment of the attack, which implies a decrease in the amplitude of the attack. The noticeable impact of this decrease is very small but can nevertheless be avoided. The figure 3 illustrates the same example as the figure 2 , in which, before smoothing, the attenuation factor value is forced to 1 for the few samples of the sub-block preceding the sub-block where the attack is located. Part (c) of the figure 3 give an example of such a correction.

In this example, the value of factor 1 has been assigned to the last 16 samples of the sub-block preceding the attack, starting from the index 364. Thus, the smoothing function progressively increases the factor to have a value close to 1 at the moment. of the attack. The amplitude of the attack is then preserved, as illustrated in part d) of the figure 3 however, some pre-echo samples are not attenuated.

In the example of the figure 3 attenuation pre-echo reduction does not reduce the pre-echo to the attack level because of gain smoothing.

This pre-echo reduction technique is however perfectible for certain types of signals such as modern music signals, for example. Indeed, in some cases, a false pre-echo detection can take place. The figure 4 illustrates an example of such an original signal, not coded so without pre-echo. This is a beat of an electronic / synthetic percussion instrument. It can be observed that before the net attack to the index 1600 there is a synthetic noise which starts towards the index 1250. This synthetic noise which is therefore part of the signal would be detected as a pre-echo by the algorithm pre-echo detection described above, assuming perfect signal coding / decoding. The pre-echo attenuation processing would therefore suppress this component of the signal. This would distort the decoded signal (when the coding / decoding is perfect), which is undesirable.

There is therefore a need for an improved decoding pre-echo discrimination and attenuation technique which makes it possible to make pre-echo detection reliable and to avoid false detections without any auxiliary information being transmitted by the encoder.

The present invention improves the state of the art.

• calcul d'un coefficient directeur des énergies pour au moins deux sous-blocs de la trame courante précédant le sous-bloc dans lequel une attaque est détectée;
• comparaison du coefficient directeur à un seuil prédéfini; et
• inhibition du traitement d'atténuation de pré-écho dans la zone de pré-écho dans le cas où le coefficient directeur calculé est inférieur au seuil prédéfini.

For this purpose, the present invention relates to a method of pre-echo discrimination and attenuation in a digital audio signal generated from a transform coding, in which, for a current frame decomposed into sub-blocks, the Low energy low sub-blocks preceding a sub-block in which a transition or attack is detected determine a pre-echo area in which a pre-echo attenuation processing is performed. The method is such that, in the case where an attack is detected from the third sub-block of the current frame, it comprises the following steps:

• calculating an energies directing coefficient for at least two sub-blocks of the current frame preceding the sub-block in which an attack is detected;
• comparing the guideline with a predefined threshold; and
• inhibiting the pre-echo attenuation processing in the pre-echo zone in the case where the calculated master coefficient is lower than the predefined threshold.

The energy director coefficient calculated for the sub-blocks preceding the position of the attack makes it possible to check the tendency of increase of the energy of the signal in the pre-echo zone. This makes reliable detection of pre-echoes by avoiding false detection of pre-echoes. By observing the figure 1 it can be seen that the pre-echo has a typical characteristic: its energy has a growing tendency in approaching the original pre-echo attack. The shape of the weighting windows of the addition-overlap explain this. Even if the pre-echo has a nearly constant energy before the overlap-addition, the signals at the input of the add-over module are multiplied by weighting windows whose weight decreases towards the past. In the case of the example signal of the figure 4 , the energy of the signal before the attack is approximately constant which makes it possible to differentiate it from a pre-echo. Thus, verification of increasing signal energy in the pre-echo area increases the reliability of the pre-echo detection.

In a particular embodiment, the method further comprises a step of decomposing the digital audio signal into at least two sub-signals according to a frequency criterion and in that the comparison calculation steps are performed for at least one of the subsignals.

When the position of the attack is detected in the third sub-block of the current frame, the energy of two sub-blocks is used in the pre-echo zone to calculate a directional coefficient and compare it to a threshold. With only two points, only the verification for the high-frequency sub-signal in the case of two sub-signal decomposition is sufficient to detect a false pre-echo detection.

In the case where the number of sub-blocks preceding the sub-block where a driving position has been detected is sufficient, the method further comprises a step of decomposing the digital audio signal into at least two sub-signals as a function of a frequency criterion and in that the calculation and comparison steps are performed for each of the sub-signals, the inhibition of the pre-echo attenuation processing in the pre-echo zone of all the sub-signals performing when a calculated master coefficient is below the predefined threshold for at least one sub-signal.

The division into sub-signals thus makes it possible to carry out a pre-echo attenuation independently and adapted in the sub-signals. The detection reliability of the pre-echo zone is enhanced for each of the sub-signals by checking the value of the respective coefficient coefficients.

According to a particular embodiment, a different threshold is defined by sub-signal.

This makes it possible to adapt the verification to the spectral characteristics of the sub-signals.

In one embodiment, the steering coefficient is calculated using a least squares estimation method.

This calculation method is of low complexity.

In one possible embodiment, the steering coefficient is normalized.

Thus the steering coefficient is more easily comparable to a threshold when it is different from 0.

In a possible embodiment, in the case where an attack is detected in the first or second sub-block of the current frame, a direction coefficient calculated for the previous frame is used for the comparison step.

• un module de calcul calculant un coefficient directeur des énergies pour au moins deux sous-blocs de la trame courante précédant le sous-bloc dans lequel une attaque est détectée;
• un comparateur apte à effectuer une comparaison du coefficient directeur à un seuil prédéfini; et
• un module de discrimination apte à inhiber le traitement d'atténuation de pré-écho dans la zone de pré-écho dans le cas où le coefficient directeur calculé est inférieur au seuil prédéfini.

The present invention also relates to a pre-echo discrimination and attenuation device in a digital audio signal generated from a transform coding, comprising a transition detection or attack module, a zone discrimination module. pre-echo and a pre-echo attenuation processing module, a pre-echo attenuation processing being performed for a current sub-block decomposed frame, in the low energy sub-blocks preceding a sub-block wherein a transition or attack is detected determining a pre-echo area. The device is such that, in the case where an attack is detected from the third sub-block of the current frame, it further comprises:

• a calculation module calculating an energies directing coefficient for at least two sub-blocks of the current frame preceding the sub-block in which an attack is detected;
• a comparator capable of making a comparison of the steering coefficient with a predefined threshold; and
• a discrimination module capable of inhibiting the pre-echo attenuation processing in the pre-echo zone in the case where the calculated directing coefficient is lower than the predefined threshold.

The advantages of this device are the same as those described for the method of discrimination and attenuation processing that it implements.

The invention relates to a decoder of a digital audio signal comprising a device as described above.

The invention also relates to a computer program comprising code instructions for implementing the steps of the method as described above, when these instructions are executed by a processor.

Finally, the invention relates to a storage medium, readable by a processor, integrated or not to the processing device, optionally removable, storing a computer program implementing a method of treatment as described above.

• la figure 1 décrite précédemment illustre un système de codage-décodage par transformée selon l'état de l'art;
• la figure 2 décrite précédemment illustre un exemple de signal audionumérique pour lequel une méthode d'atténuation selon l'état de l'art est effectuée;
• la figure 3 illustre un autre exemple de signal audionumérique pour lequel une méthode d'atténuation selon l'état de l'art est effectuée;
• la figure 4 décrite précédemment illustre un exemple d'un signal pour lequel la technique de l'état d'art détecterait à tort un pré-écho;
• la figure 5 illustre un mode de réalisation d'un procédé et d'un dispositif de discrimination et de traitement d'atténuation de pré-écho compris dans un décodeur selon l'invention;
• la figure 6 illustre un exemple de fenêtres d'analyse et de fenêtres de synthèse à faible retard pour le codage et le décodage par transformée susceptible de créer le phénomène de pré-écho;
• la figure 7 illustre un exemple de signal audionumérique pour lequel la méthode d'atténuation de pré-écho selon un mode de réalisation de l'invention est mis en oeuvre;....
• la figure 8 illustre un exemple matériel de dispositif de discrimination et de traitement d'atténuation selon l'invention.

Other features and advantages of the invention will appear more clearly on reading the following description, given solely by way of nonlimiting example, and with reference to the appended drawings, in which:

• the figure 1 previously described illustrates a state-of-the-art transform coding-decoding system;
• the figure 2 described above illustrates an example of a digital audio signal for which a mitigation method according to the state of the art is performed;
• the figure 3 illustrates another example of a digital audio signal for which an attenuation method according to the state of the art is performed;
• the figure 4 described above illustrates an example of a signal for which the state of the art technique would erroneously detect a pre-echo;
• the figure 5 illustrates an embodiment of a method and a device for discriminating and pre-echo attenuation processing included in a decoder according to the invention;
• the figure 6 illustrates an example of low-delay analysis windows and synthesis windows for transform coding and decoding that can create the pre-echo phenomenon;
• the figure 7 illustrates an example of a digital audio signal for which the pre-echo attenuation method according to one embodiment of the invention is implemented;
• the figure 8 illustrates a hardware example of discrimination device and attenuation processing according to the invention.

With reference to the figure 5 A discrimination device 600 and pre-echo reduction processing is described. The attenuation processing device 600 as described below is included in a decoder comprising a reverse quantization module 610 (Q ^-1 ) receiving a signal S, a reverse transformation module 620 (MDCT ^-1 ), a module 630 of addition / recap signal reconstruction as described with reference to FIG. figure 1 and delivering a reconstructed signal x _rec ( n ) to the attenuation discrimination and processing device according to the invention. It may be noted that here we take the example of the MDCT transform which is the most common in speech and audio coding, however the device 600 also applies to any other type of transform (FFT, DCT, etc.).

At the output of the device 600, a processed signal Sa is provided in which a pre-echo attenuation has been performed.

The device 600 implements a discrimination and pre-echo attenuation processing method in the decoded signal x _rec ( n ) .

In one embodiment of the invention, the discrimination and attenuation processing method includes a step of detecting (E601) attacks that may generate a pre-echo, in the decoded signal x _rec ( n ) .

Thus, the device 600 comprises a detection module 601 able to implement a step of detecting (E601) the position of an attack in a decoded audio signal.

An onset (or onset ) is a rapid transition and a sudden change in the dynamics (or amplitude) of the signal. This type of signal may be referred to by the more general term "transient". In the following and without loss of generality, we will use only the terms of attack or transition to designate also transients.

Each current frame of L samples of the decoded signal x _rec ( n ) is divided into K sub-blocks of length L ', with for example L = 640 samples (20 ms) at 32 kHz, L' = 80 samples (2.5 ms) and K = 8. Preferably the size of these sub-blocks is therefore identical but the invention remains valid and easily generalizable when the sub-blocks have a variable size. This can be the case for example when the length of the frame L is not divisible by the number of sub-blocks K or if the frame length is variable.

Special low-delay analysis-synthesis windows similar to those described in ITU-T G.718 are used for the analysis part and for the synthesis part of the MDCT transformation. An example of such windows is illustrated with reference to the figure 6 . The delay generated by the transformation is only 280 samples in contrast to the delay of 640 samples in the case of use of conventional sinusoidal windows. Thus the MDCT memory with special low-delay analysis-synthesis windows contains only 140 independent samples (not folded with the current frame) unlike the 320 samples in the case of using conventional sinusoidal windows.

We can indeed notice on the figure 6 for the analysis windows (Ana.), that the folding zone is limited by the dashed lines between the samples 820 and 1100. The folding line is shown in dashed line with the sample 960.

For the synthesis (Synth.), Only the samples represented by the interval M (140 samples) are necessary to obtain the information on the folding area of the analysis, exploiting the symmetry. These samples contained in memory are then useful for decoding this folding zone by also using the folded samples of the window of the next frame. In the case of an attack in this zone between the samples 820 and 1100, the average energy of the samples represented by the interval M is clearly greater than the energy of subframes preceding the sample 820. The abrupt increase of the energy of the interval M contained in the MDCT memory can therefore signal an attack in the next frame which can generate a pre-echo in the current frame.

The MDCT x _MDCT ( n ) memory is used which gives a time-folding version of the future signal. With the special low-delay analysis-synthesis windows as illustrated in figure 6 , we retain only one (K '= 1) block length L _m (0) = 140 which contains all the independent samples of the MDCT memory. Despite the higher number of samples in this sub-block, its energy remains comparable with that of the sub-blocks of the current frame (if the signal remains stable) because the memory part has been windowed (thus attenuated) by the window of analysis.

Indeed, the figure 1 shows that the pre-echo influences the frame that precedes the frame where the attack is located, and it is desirable to detect an attack in the future frame which is partly contained in the MDCT memory.

quand le k-ième sous-bloc se situe dans la trame courante et, comme: $$\mathit{En}\left(k\right)={\displaystyle \sum _{n=0}^{L_{\mathit{mem}}-1}{x}_{\mathit{MDCT}}{\left(n\right)}^2}$$

quand le sous-bloc est dans la mémoire MDCT (qui représente le signal disponible pour la trame future) et L_mem est la longueur du sous-bloc de la partie mémoire :
L'énergie moyenne des sous-blocs dans la trame courante s'obtient donc comme : $$\overline{\mathit{En}}=\frac{1}{K}{\displaystyle \sum _{k=0}^{K-1}\mathit{En}\left(k\right)}$$

On définit également l'énergie moyenne des sous-blocs dans la deuxième partie de la trame courante comme (supposant que K est un nombre pair): $$\overline{\mathit{En}}\text{'}=\frac{2}{K}{\displaystyle \sum _{k=K/2}^{K-1}\mathit{En}\left(k\right)}$$

The current frame and the MDCT memory can be seen as concatenated signals forming a signal cut into (K + K ') consecutive sub-blocks. Under these conditions, the energy in the k-th sub-block is defined as: $$\mathit{In}\left(k\right)={\displaystyle \underset{\mathrm{not}=\mathit{kL}\text{'}}{\overset{\left(k+1\right)\mathrm{The}\text{'}-1}{\Sigma }}{x}_{\mathit{rec}}{\left(\mathrm{not}\right)}^2,k=0...,K-1}$$

when the k-th sub-block is in the current frame and, like: $$\mathit{In}\left(k\right)={\displaystyle \underset{\mathrm{not}=0}{\overset{{\mathrm{The}}_{\mathit{same}}-1}{\Sigma }}{x}_{\mathit{MDCT}}{\left(\mathrm{not}\right)}^2}$$

when the sub-block is in the MDCT memory (which represents the signal available for the future frame) and L _mem is the length of the sub-block of the memory part:
The average energy of the sub-blocks in the current frame is thus obtained as: $$\widetilde{\mathit{In}}=\frac{1}{K}{\displaystyle \underset{k=0}{\overset{K-1}{\Sigma }}\mathit{In}\left(k\right)}$$

The average energy of the sub-blocks in the second part of the current frame is also defined as (assuming that K is an even number): $$\widetilde{\mathit{In}}\text{'}=\frac{2}{K}{\displaystyle \underset{k=K/2}{\overset{K-1}{\Sigma }}\mathit{In}\left(k\right)}$$

dépasse un seuil prédéfini, dans un des sous-blocs considérés. D'autres critères de détection de pré-écho sont possibles sans changer la nature de l'invention.
Par ailleurs, on considère que la position de l'attaque est définie comme $$\mathit{pos}=\min \left(L\text{'}.\left(arg\mathrm{}\underset{k=\mathrm{0,}K+K\text{'}-1}{\max }\left(\mathit{En}\left(k\right)\right)\right),L\right)$$

où la limitation à L assure que la mémoire MDCT n'est jamais modifiée. D'autres méthodes d'estimation plus précise de la position de l'attaque sont également possibles.An attack associated with a pre-echo is detected if the report $R\left(k\right)=\frac{\underset{\mathrm{not}=0K+K\text{'}-1}{\max }\left(\mathit{In}\left(\mathrm{not}\right)\right)}{\mathit{In}\left(k\right)}$

exceeds a predefined threshold, in one of the sub-blocks considered. Other pre-echo detection criteria are possible without changing the nature of the invention.
Moreover, we consider that the position of the attack is defined as $$\mathit{pos}=\min \left(\mathrm{The}\text{'}.\left(\mathrm{at}r\mathrm{boy\; Wut}\mathrm{}\underset{k=0K+K\text{'}-1}{\max }\left(\mathit{In}\left(k\right)\right)\right),\mathrm{The}\right)$$

where the limitation to L ensures that the MDCT memory is never changed. Other methods for more accurate estimation of the attack position are also possible.

The device 600 also comprises a pre-echo zone discrimination module 602 implementing a step of determining (E602) a pre-echo zone (ZPE) preceding the detected driving position. Pre-echo zone is here referred to as the zone covering the samples before the estimated position of the attack which are disturbed by the pre-echo generated by the attack and where attenuation of this pre-echo is desirable. In the embodiment shown the pre-echo area can be determined on the decoded signal.

In one embodiment of obtaining the pre-echo zones, the energies En ( k ) are concatenated in chronological order, with the time envelope of the decoded signal first, then the envelope of the signal of the following estimated frame from the memory of the MDCT transform. According to this concatenated temporal envelope and mean energies In and In of the previous frame, the presence of pre-echo is detected for example if the ratio R ( k ) exceeds a threshold, typically this threshold is 16.

The sub-blocks in which a pre-echo has been detected thus constitute a pre-echo zone, which in general covers the samples n = 0, ···, pos -1, ie from the beginning of the current frame to the position of the attack ( pos ) . It can also be noted that the pre-echo zone may well extend over the entire current frame if the attack has been detected in the future frame.

The device 600 comprises a calculation module 603 capable of implementing a step of calculating a director coefficient (or variation trend indicator) of the energies of the sub-blocks preceding the sub-block in which an attack has been detected.

We define the linear model which represents a set of n realizations (t _i , e _i ), 0 <= i <n where t _i are the temporal indices of the sub-blocks and e _i their energies, with the equation $$e={\mathrm{<figure-callout\; id="0"\; label="b"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">b</figure-callout>}}_0+b_{1}t$$

Where b ₀ is the value at time t = 0 and b ₁ is the coefficient of direction. The guideline gives information on the (mean) trend of energy change. A positive directing coefficient signals an increase in energies. A value close to 0 indicates constant energy.

We can determine the value of b ₁ for example by linear regression according to the least squares: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_1=\frac{{\displaystyle \Sigma \left(t_i-\tilde{t}\right)\left(e_i-\tilde{e}\right)}}{{\displaystyle \Sigma {\left(t_i-\tilde{t}\right)}^2}}$$

Where the summation is performed on predetermined indices i.

The value of b ₁ also depends on the magnitude (in absolute value) of the energies; it is indeed homogeneous with energy over time. To be able to better compare the value of b ₁ with a threshold (for example fixed) one can suppress this dependency. For example, the value of b ₁ can be divided by the mean value of the energies to obtain the standardized guideline: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathrm{not}}=\frac{{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_1\mathrm{not}}{{\displaystyle \Sigma e_i}}$$

Alternatively we can take the correlation coefficient. $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathrm{not}\_\mathit{alt}}=\frac{{\displaystyle \Sigma \left(t_i-\tilde{t}\right)\left(e_i-\tilde{e}\right)}}{\sqrt{{\displaystyle \Sigma {\left(t_i-\tilde{t}\right)}^2}{\displaystyle \Sigma {\left(e_i-\widetilde{\stackrel{~}{e}}\right)}^2}}}$$

This alternative solution has a higher computational complexity because it requires calculating a square root.

Other methods of estimating the directing coefficient are also possible, such as the median-median Tukey method.

It can also be noted that when the reference coefficient has to be compared with a threshold of zero value - which amounts to verifying the sign of this coefficient - it is not necessary to standardize this coefficient.

$$b_1<\mathit{seuil}.\frac{{\displaystyle \sum e_i}}{n}$$

On the other hand, instead of normalizing the steering coefficient, it will be possible to make the threshold variable because the following relations are equivalent: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathrm{not}}=\frac{{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_1\mathrm{not}}{{\displaystyle \Sigma e_i}}<\mathit{<figure-callout\; id="1"\; label="threshold\; b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">threshold\; </figure-callout>}$$

$$b_{}$$$${}_1<\mathit{threshold}.\frac{{\displaystyle \Sigma e_i}}{\mathrm{not}}$$

If the attack is detected in the first or second sub-block verification according to the invention is not possible. If the attack is detected in the third sub-block one has the energy of 2 sub-blocks in the pre-echo zone, e ₀ and e ₁ to make this verification (e ₁ being the closest to the attack). With 2 points equation (3) is simplified as follows: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathrm{not}}=\frac{2\left(e_1-e_0\right)}{{\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">e</figure-callout>}}_1+{\mathrm{<figure-callout\; id="0"\; label="e"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">e</figure-callout>}}_0}$$

If the attack is detected in the fourth sub-block we have the energy of 3 sub-blocks in the pre-echo zone, e ₀ , e ₁ and e ₂ to make this verification (e ₂ being the closest of the attack). With 3 points equation (3) is simplified as follows: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathrm{not}}=\frac{3\left(e_2-{\mathrm{<figure-callout\; id="0"\; label="e"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">e</figure-callout>}}_0\right)}{2\left(e_2+{\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">e</figure-callout>}}_1+{\mathrm{<figure-callout\; id="0"\; label="e"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">e</figure-callout>}}_0\right)}$$

If we have 4 or more sub-blocks we can calculate the guideline on 4 or more sub-blocks. Experience shows that the verification of the coefficient calculated on the 3 sub-blocks preceding the sub-block where the attack was detected is sufficient to avoid false detections of the pre-echoes - this conclusion applies for the case of 8 sub-blocks on each frame of 20 ms and can be adapted according to the size of the sub-blocks and the frame.

Thus, in the preferred embodiment, the steering coefficient is calculated with at most 3 sub-blocks. This makes it possible to limit the maximum complexity of the calculation of the steering coefficient.

According to the invention, the normalized standard coefficient b _1n thus obtained is then compared with the step E604 by a comparator module 604 at a predefined threshold. The threshold may be predefined to a fixed value or may be variable depending for example on the classification of the signal according to a speech or music criterion. Typically this threshold is equal to 0 if we only check that the energy does not decrease or equal to 0.2 if we impose a slight increase in energy in the pre-echo zone. If the standardized guideline b _1n is below this threshold it is concluded that the signal in the pre-echo zone does not correspond to a typical pre-echo and the pre-echo attenuation in this zone at step E602 is inhibited. Thus, it is avoided that a decoded signal whose original input signal contains a low energy component before an attack is erroneously modified / altered by the pre-echo attenuation module by detecting that component as a pre-echo.

A pre-echo attenuation is implemented in step E607 by the attenuation module 607 for the discriminated pre-echo area. The attenuation factor is for example calculated as in the request FR 08 56248 . In the case where the module 604 has detected a false pre-echo detection, the attenuation factor can be forced to 1 thus inhibiting the attenuation or the discrimination module 602 does not discriminate this zone as a pre-echo zone. echo, the attenuation module is not requested.

In a particular embodiment, the device 600 further comprises a signal decomposition module 605, able to perform a step E605 of decomposing the decoded signal into at least two sub-signals according to a predetermined criterion. This method is notably described in the application FR12 62598 which we recall here some elements.

• le premier sous-signal x _rec,ss1(n) est obtenu par filtrage passe bas en utilisant un filtre FIR (filtre à réponse impulsionnelle finie) à 3 coefficients et à phase nulle de fonction de transfert c(n)z ^-1+(1-2c(n))+c(n)z avec c(n) une valeur comprise entre 0 et 0.25, où [c(n),1-2c(n),c(n)] sont les coefficients du filtre passe bas ; ce filtre est mis en oeuvre avec l'équation aux différences : $$x_{\mathit{rec},\mathit{ss}1}\left(n\right)=c\left(n\right)x_{\mathit{rec}}\left(n-1\right)+\left(1-2c\left(n\right)\right)x_{\mathit{rec}}\left(n\right)+c\left(n\right)\mathrm{<figure-callout\; id="1"\; label="x\; n\; +"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">x</figure-callout>}\left(n+\right)\left(1\right)$$
  

In a particular embodiment of the invention, the decoded signal x _rec ( n ) is decomposed in step E605 into two sub-signals as follows:

• the first sub-signal x _{rec, ss 1} ( n ) is obtained by low-pass filtering using a FIR filter (finite impulse response filter) with 3 coefficients and a null transfer function phase c ( n ) z ^-1 + (1- 2c ( n )) + c ( n ) z with c ( n ) a value between 0 and 0.25, where [ c ( n ) , 1-2 c ( n ) , c ( n )] are the coefficients the low pass filter; this filter is implemented with the difference equation: $$x_{\mathit{rec},\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}\left(\mathrm{not}\right)=\mathrm{vs}\left(\mathrm{not}\right)x_{\mathit{rec}}\left(\mathrm{not}-1\right)+\left(1-2\mathrm{vs}\left(\mathrm{not}\right)\right)x_{\mathit{rec}}\left(\mathrm{not}\right)+\mathrm{vs}\left(\mathrm{not}\right)x\left(\mathrm{not}+1\right)$$
  

In a particular embodiment, a constant value c ( n ) = 0.25 is used.

• le deuxième sous signal x _rec,ss2(n) est obtenu par filtrage passe haut complémentaire en utilisant un filtre FIR à 3 coefficients et à phase nulle de fonction de transfert -c(n)z ^-1+2c(n)-c(n)z, où [-c(n),2c(n),-c(n)] sont les coefficients du filtre passe haut ; ce filtre est mis en oeuvre avec l'équation aux différences : x _rec,ss2(n)=-c(n)x_rec (n-1)+2c(n)x_rec (n)-c(n)x(n+1). Le sous signal x _rec,ss2(n) résultant de ce filtrage, contient donc des composantes plutôt hautes fréquences du signal décodé.

It may be noted that the sub-signal x _{rec, s1} ( n ) resulting from this filtering, therefore contains rather low-frequency components of the decoded signal.

• the second sub-signal x _{rec, ss 2} ( n ) is obtained by complementary high-pass filtering using a FIR filter with 3 coefficients and with a zero transfer function phase -c ( n ) z ^-1 + 2 c ( n ) - c ( n ) z , where [- c ( n ) , 2c ( n ), - c ( n )] are the coefficients of the high pass filter; this filter is implemented with the difference equation: x _{rec, ss 2} ( n ) = -c ( n ) x _rec ( n -1) + 2 c ( n ) x _rec ( n ) -c ( n ) x ( n +1). The sub-signal x _{rec, ss 2} ( n ) resulting from this filtering, therefore contains rather high-frequency components of the decoded signal.

Note that x _{rec, ss 1} ( n ) + x _{rec, ss 2} ( n ) = x _rec ( n ).

It is therefore also possible to obtain x _{rec, ss 2} ( n ) by subtracting x _{rec, ss 1} ( n ) from x _rec ( n ), which reduces the computational complexity: x _{rec, ss 2} ( n ) = x _rec ( n ) - x _{rec, ss 1} ( n )

The combination of the attenuated sub-signals for obtaining the attenuated signal Sa is made by simply adding the attenuated sub-signals to the step E608 described later.

x _rec,ss2(n) est toujours calculé comme x _rec,ss2(n) = x_rec (n)-x _rec,ss1(n). In order not to use a future signal for these filterings, it is possible, for example, to complete the signal decoded by a sample at 0 at the end of the block. In the case of the decoded signal completed by a sample at 0 at the end of the block for n = L-1, the sub-signal x _{rec, ss 1} ( n ) is obtained by: $$x_{\mathit{rec},\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}\left(\mathrm{The}-1\right)=\mathrm{vs}\left(\mathrm{The}-1\right)x_{\mathit{rec}}\left(\mathrm{The}-2\right)+\left(1-2\mathrm{vs}\left(\mathrm{The}-1\right)\right)x_{\mathit{rec}}\left(\mathrm{The}-1\right),$$

x _{rec, ss 2} ( n ) is always calculated as x _{rec, ss 2} ( n ) = x _rec ( n ) - x _{rec, ss 1} ( n ) .

It can be noted that the two sub-signals remain here at the same sampling frequency as the decoded signal.

A step E606 for calculating pre-echo attenuation factors is implemented in the calculation module 606. This calculation is done separately for the two sub-signals.

These attenuation factors are obtained by sampling the pre-echo zone determined in E602 as a function of the frame in which the attack was detected and the previous frame.

We then obtain the factors g _{press 1} '( n ) and g _{pre, ss 2} ' ( n ) where n is the index of the corresponding sample. These factors will eventually be smoothed to obtain the factors g _{pre, ss 1} ( n ) and g _{pre, ss 2} ( n ) respectively. This smoothing is especially important for sub-signals containing the low-frequency components (so for g _{pre, ss 1} '( n ) in this example).

An embodiment of attenuation calculation is described in the patent application FR 08 56248 . The attenuation factors are calculated by sub-block. In the method described here, they are additionally calculated separately for each sub signal. For samples preceding the detected attack, the attenuation factors g _{pre, ss 1} '( n ) and g _{pre, ss 2} ' ( n ) are thus calculated. Then these attenuation values are optionally smoothed to obtain the attenuation values per sample.

où f est une fonction décroissante à valeurs entre 0 et 1, par exemple f=0 si R(k) <= 16, f = 0.1 si 16 > R(k) >= 32 et f=0.01 si r(k) >32.The calculation of the attenuation factor of a sub-signal (for example g _{pre, ss 2} '( n )) can be similar to that described in the patent application. FR 08 56248 for the decoded signal according to the ratio R ( k ) (also used for the detection of the attack) between the energy of the sub-block of higher energy and the energy of the k-th sub-block of the decoded signal. We initialize g _{pre, ss 2} '( n ) as: $${\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\text{'}\left(\mathrm{not}\right)=\mathrm{boy\; Wut}\left(k\right)=f\left(R\left(k\right)\right),\mathrm{}\mathrm{not}=\mathit{kL}\text{'},...,\left(k+1\right)\mathrm{The}\text{'}-1;\mathrm{}k=0...,K-1$$

where f is a decreasing function with values between 0 and 1, for example f = 0 if R (k) <= 16, f = 0.1 if 16> R (k)> = 32 and f = 0.01 if r (k)> 32.

If the variation of the energy with respect to the maximum energy is small, then no attenuation is necessary. The factor is then set to a attenuation value that inhibits the attenuation, that is to say 1. Otherwise, the attenuation factor is between 0 and 1. This initialization can be common for all the sub-signals. .

The attenuation values are then refined by sub-signal to be able to adjust the optimal sub-signal attenuation level based on the characteristics of the decoded signal. For example, the attenuations can be limited according to the average energy of the sub-signal of the previous frame because it is not desirable that after the pre-echo attenuation processing, the signal energy becomes less than the average energy per sub-block of the signal preceding the processing zone (typically that of the previous frame or that of the second half of the previous frame).

On connait également par mémorisation l'énergie moyenne de la trame précédente En _ss2 et celle de la deuxième moitié de la trame précédente En _ss2' qui peuvent être calculés (à la trame précédente) comme : $$\overline{E{n}_{\mathit{ss}2}}=\frac{1}{K}{\displaystyle \sum _{k=0}^{K-1}E{n}_{\mathit{ss}2}\left(k\right)}$$

et $$\overline{E{n}_{\mathit{ss}2}}\text{'}=\frac{2}{K}{\displaystyle \sum _{k=K/2}^{K-1}E{n}_{\mathit{ss}2}\left(k\right)}$$

où les indices de sous-bloc de 0 à K correspondent à la trame courante.This limitation can be made in a manner similar to that described in the patent application FR 08 56248 . For example, for the second sub-signal x _{rec, ss 2} ( n ), the energy in the sub-block K of the current frame is first calculated as: $$E{\mathrm{not}}_{\mathit{ss}2}\left(k\right)={\displaystyle \underset{\mathrm{not}=\mathit{kL}\text{'}}{\overset{\left(k+1\right)\mathrm{The}\text{'}-1}{\Sigma }}{x}_{\mathit{rec},\mathit{ss}2}{\left(\mathrm{not}\right)}^2,k=0...,K-1}$$

We also know by memorization the average energy of the previous frame In _{ss 2} and that of the second half of the previous frame In _{ss 2} 'which can be calculated (at the previous frame) as: $$\widetilde{E{\mathrm{not}}_{\mathit{ss}2}}=\frac{1}{K}{\displaystyle \underset{k=0}{\overset{K-1}{\Sigma }}E{\mathrm{not}}_{\mathit{ss}2}\left(k\right)}$$

and $$\widetilde{E{\mathrm{not}}_{\mathit{ss}2}}\text{'}=\frac{2}{K}{\displaystyle \underset{k=K/2}{\overset{K-1}{\Sigma }}E{\mathrm{not}}_{\mathit{ss}2}\left(k\right)}$$

where the sub-block indices from 0 to K correspond to the current frame.

où l'énergie moyenne du segment précédent est approximée par max ( En _ss2 , En _ss2').For the sub-block k to be processed, the limit value of lim _{g, ss 2} ( k ) can be calculated in order to obtain exactly the same energy as the average energy per sub-block of the segment preceding the sub-block to be processed. . This value is of course limited to a maximum of 1 since we are interested here in the attenuation values. More precisely : $${\lim }_{\mathrm{boy\; Wut},\mathit{ss}2}\left(k\right)=\min \left(\sqrt{\frac{\max \left(\widetilde{E{\mathrm{not}}_{\mathit{ss}2}},\widetilde{E{\mathrm{not}}_{\mathit{ss}2}}\text{'}\right)}{E{\mathrm{not}}_{\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}2}\left(k\right)}\mathrm{,\; 1}}\right)$$

where the average energy of the previous segment is approximated by max ( In _{ss 2} , In _{ss 2} ').

The value lim _{g , ss 2} ( k ) thus obtained serves as a lower limit in the final calculation of the attenuation factor of the sub-block: $${\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\text{'}\left(\mathrm{not}\right)=\max \left({\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\text{'}\left(\mathrm{not}\right),{\lim }_{\mathrm{boy\; Wut},\mathit{ss}2}\left(k\right)\right),\mathrm{}\mathrm{not}=\mathit{kL}\text{'},...,\left(k+1\right)\mathrm{The}\text{'}-1;\mathrm{}k=0...,K-1$$

Les atténuations associées aux échantillons du sous-bloc de l'attaque sont toutes mises à 1 même si l'attaque se situe vers la fin de ce sous-bloc.In a first variant embodiment the pre-echo zone where the attenuation extends from the beginning of the current frame to the beginning of the sub-block in which the attack has been detected - up to the index pos where $\mathit{pos}=\min \left(\mathrm{The}\text{'}.\left(\mathit{arg}\underset{k=0K+K\text{'}-1}{\max }\left(\mathit{In}\left(k\right)\right)\right),\mathrm{The}\right).$

The attenuations associated with the sub-block samples of the attack are all set to 1 even if the attack is towards the end of this sub-block.

In another variant embodiment, the start position of the attack pos is refined in the sub-block of the attack, for example by cutting the sub-block into sub-sub-blocks and observing the evolution of the energy of these sub-sub-blocks. Suppose that the position of the beginning of the attack is detected in the sub-block k, k> 0 and the beginning of the refined attack pos is in this sub-block, the attenuation values for the samples of this sub-block -block which are before the index pos can be initialized according to the attenuation value corresponding to the last sample of the previous sub-block: $${\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\text{'}\left(\mathrm{not}\right)={\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\text{'}\left(\mathit{kL}\text{'}-1\right),\mathrm{}\mathrm{not}=\mathit{kL}\text{'},...,\mathit{pos}-1$$

All attenuations from the index pos are set to 1.

For the first sub-signal containing the low frequency components of the decoded signal, the calculation of the attenuation values based on the sub-signal x _{rec, ss 1} ( n ) may be similar to the calculation of the attenuation values by based on the decoded signal x _rec ( n ) . Thus, in an alternative embodiment, for the sake of reducing calculation complexity, the attenuation values can be determined based on the decoded signal x _rec ( n ) . In the case where the detection of attacks is made on the decoded signal, it is no longer necessary to recalculate energies of the sub-blocks because for this signal the energy values by sub-block are already calculated to detect the attacks. As for the great majority of the signals the low frequencies are much more energetic than the high frequencies, the energies by sub-block of the decoded signal x _rec ( n ) and of the sub-signal x _{rec, ss 1} ( n ) are very close, this approximation gives a very satisfactory result.

The attenuation factors g _{pre , ss1} ( n ) and g _{pre , ss 2} ( n ) determined by sub-blocks can then be smoothed by an applied smoothing function sample by sample to avoid abrupt changes in the attenuation factor at block boundaries. This is particularly important for sub-signals containing low frequency components such as sub-signal x _{rec, ss 1} ( n ) but not necessary for sub-signals containing only high frequency components such as sub-signal x _{rec, ss 2} ( n ) .

The figure 7 illustrates an example of applying an attenuation gain with smoothing functions represented by the L arrows.

This figure illustrates in a), an example of an original signal, in b), the decoded signal without pre-echo attenuation, in c), the attenuation gains for the two sub-signals obtained according to the decomposition step E605 and in d), the decoded signal with pre-echo attenuation of steps E607 and E608 (i.e. after combining the two attenuated sub-signals).

It may be noted in this figure that the attenuation gain represented in dashed line and corresponding to the gain calculated for the first sub-signal comprising low frequency components, comprises smoothing functions as described above. The attenuation gain represented in solid line and calculated for the second sub-signal comprising high frequency components, does not include smoothing gain.

avec la convention que g _pre,ss1'(n) n = -(u-1),···,-1 sont les derniers u-1 facteurs d'atténuation obtenus pour les derniers échantillons du sous-bloc précédent du sous-signal x _rec,ss1(n). Typiquement u = 5 mais une autre valeur pourrait être utilisée. En fonction du lissage utilisé, la zone de pré-écho (le nombre des échantillons atténués) peut donc être différente pour les 2 sous-signaux traités séparément, même si la détection de l'attaque est faite en commun sur la base du signal décodé.The signal represented in d) clearly shows that the pre-echo has been effectively attenuated by the attenuation processing implemented.
The smoothing function is for example preferably defined by the following equations: $${\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}\left(\mathrm{not}\right)=\frac{1}{u}{\displaystyle \underset{i=0}{\overset{u-1}{\Sigma }}{\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}1}\text{'}}\left(\mathrm{not}-i\right),\mathrm{}\mathrm{not}=0\cdots ,\mathrm{The}-1$$

with the convention that g _{pre, ss 1} '( n ) n = - ( u -1), ···, -1 are the last u-1 attenuation factors obtained for the last samples of the previous sub-block of the sub -signal x _{rec, ss 1} ( n ). Typically u = 5 but another value could be used. Depending on the smoothing used, the pre-echo zone (the number of attenuated samples) may therefore be different for the 2 sub-signals processed separately, even if the detection of the attack is made in common on the basis of the decoded signal. .

The smoothed attenuation factor does not go back to 1 at the time of the attack, which implies a decrease in the amplitude of the attack. The noticeable impact of this decrease is very small but must nevertheless be avoided. To alleviate this problem the attenuation factor value can be forced to 1 for the u-1 samples preceding the index pos where the onset of the attack is. This is equivalent to advancing the pos marker of u-1 samples for the sub signal where the smoothing is applied. Thus the smoothing function gradually increases the factor to have a value 1 at the time of the attack. The amplitude of the attack is then preserved.

In this embodiment with signal decomposition, the verification of the increase of the energy of the pre-echo zone according to the invention is carried out for at least one sub-signal or for each of these sub-signals.

The comparison threshold used may be different depending on the sub-signals and the number of sub-blocks available before the attack.

If, in at least one sub-signal, the normalized steering coefficient b _1n is less than the threshold of this sub-signal, the pre-echo attenuation is inhibited for all the sub-signals.

In the case of a pre-echo in a signal coming from an inverse MDCT transform, the energy of the pre-echo component increases or is at least stable in all the sub-signals. Inhibition of pre-echo processing can be done for example by setting the attenuation factors to 1 or not discriminating the area as a pre-echo zone, the module of pre-echo attenuation processing is not then requested as illustrated by way of example in the embodiment of the figure 5 by the link between block 604 and 602.

In variants, the attenuation will be inhibited separately for each sub-signal as soon as the normalized steering coefficient b _1n is lower than the threshold of this sub-signal. The inhibition may for example be implemented by setting the attenuation factors to 1 or by not soliciting the pre-echo module for the sub-signal considered.

Thus, in the particular embodiment described above with decomposition into two sub-signals, if the number of sub-blocks before the attack makes it possible to carry out this verification, the evolution of the two sub-signals is checked in both sub-signals. energy of the sub-blocks preceding the sub-block where the attack was detected, by linear regression. This verification can be done according to the steps E603 and E604, at any time after the division of the decoded signal into sub-signals (E605) and before the application of the pre-echo attenuation factors (E607). Verification is possible if at least two sub-blocks precede the sub-block where the attack was detected. If the attack is detected in the first or second sub-block verification according to the invention is not possible.

In variants, it will be possible to re-use the coefficient (s) director (s) possibly calculated (s) in the previous frame if the attack is detected in the first or second sub-block of the current frame.

If the attack is detected in the third sub-block then we have the energy of two sub-blocks in the pre-echo zone to do this verification. By experience, with two points, the verification is not sufficiently reliable in the low frequency sub-signal x _{rec, ss 1} ( n ) . We then check only the high-frequency sub-signal x _{rec, ss 2} ( n ) and only that the energy does not decrease. The high-frequency sub-signal coefficient x _{rec, ss 2} ( n ) is compared with the value threshold 0. Only its sign is important here, it is not necessary to make a normalization. It is therefore sufficient to calculate in step E603 a simple director coefficient (without normalization) such as: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathit{ss}2}=E{\mathrm{not}}_{\mathit{ss}2}\left(1\right)-E{\mathrm{not}}_{\mathit{ss}2}\left(0\right)$$

If b _1ss2 is less than 0, the attenuation of pre-echoes for this pre-echo zone is inhibited for all the sub-signals.

If the attack is detected in the fourth sub-block or a sub-block of index higher than 4, one checks the evolution of the energy of the last 3 sub-blocks in the pre-echo zone preceding the sub-block block where the attack was detected. The reference coefficient of the low-frequency sub-signal x _{rec, ss 1} ( n ) is compared to 0, only its sign is important and it is not necessary to normalize this coefficient. It is therefore sufficient to calculate a simple coefficient of direction. If the attack has been detected in the sub-block of the index id with id > = 3, this coefficient is determined as: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}=\mathit{In}\left(\mathit{id}-1\right)-E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-3\right)$$

If b _1ss1 is less than 0, pre-echo attenuation is inhibited for this pre-echo zone, and for all the sub-signals.

The direction coefficient of the high-frequency sub-signal x _{rec, ss 2} ( n ) is compared with a threshold of value 0.2. The standard guideline is calculated. If the attack has been detected in the sub-block of the index id with id > = 3, this coefficient is determined as: $${\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">b</figure-callout>}}_{1\mathit{nss}2}=\frac{3\left(E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)-E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)\right)}{2\left(E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-3\right)\right)}$$

If b _1nss2 is less than 0.2, pre-echo attenuation is inhibited for this pre-echo zone, and for all the sub-signals.

est équivalente à $$E{n}_{\mathit{ss}2}\left(\mathit{id}-1\right)-E{n}_{\mathit{ss}2}\left(\mathit{id}-2\right)<\frac{1}{7.5}\left(E{n}_{\mathit{ss}2}\left(\mathit{id}-1\right)+E{n}_{\mathit{ss}2}\left(\mathit{id}-2\right)+E{n}_{\mathit{ss}2}\left(\mathit{id}-3\right)\right)$$

évitant ainsi une opération de division pour réduire la complexité et pour faciliter la mise en oeuvre sur un processeur DSP (pour "Digital Signal Processor") à arithmétique à virgule fixe.Note that the condition $$\frac{3\left(E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)-E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)\right)}{2\left(E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-3\right)\right)}<0.2$$

is equivalent to $$E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)-E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)<\frac{1}{7.5}\left(E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-1\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-2\right)+E{\mathrm{not}}_{\mathit{ss}2}\left(\mathit{id}-3\right)\right)$$

thus avoiding a division operation to reduce the complexity and to facilitate the implementation on a DSP (for "Digital Signal Processor") to fixed point arithmetic.

The module 607 of the device 600 of the figure 5 implements the pre-echo attenuation step E607 in the pre-echo area of each of the sub-signals by applying to the subsignals of the thus calculated attenuation factors.

The pre-echo attenuation is therefore done independently in the sub-signals. Thus, in the sub-signals representing different frequency bands, the attenuation can be chosen according to the spectral distribution of the pre-echo.

Finally, a step E608 of the obtaining module 608 makes it possible to obtain the attenuated output signal (the decoded signal after pre-echo attenuation) by combining (in this example by simple addition) the attenuated sub-signals, according to the equation: $$x_{\mathit{ref},f}\left(\mathrm{not}\right)={\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}\left(\mathrm{not}\right)x_{\mathit{rec},\mathit{<figure-callout\; id="1"\; label="ss"\; filenames="imgf0006.png,imgf0007.png"\; state="\{\{state\}\}">ss</figure-callout>}1}\left(\mathrm{not}\right)+{\mathrm{boy\; Wut}}_{\mathit{pre},\mathit{ss}2}\left(\mathrm{not}\right)x_{\mathit{rec},\mathit{ss}2}\left(\mathrm{not}\right),\mathrm{}\mathrm{not}=0\cdots ,\mathrm{The}-1$$

Unlike a conventional subband decomposition, it can be noted here that the filtering used is not associated with sub-signal decimation operations and the complexity and delay ("lookahead" or future frame) are reduced to a minimum.

An exemplary embodiment of a discrimination and attenuation processing device according to the invention is now described with reference to the figure 8 .

Materially, this device 100 in the sense of the invention typically comprises a μP processor cooperating with a memory block BM including a storage and / or working memory, and a memory buffer MEM mentioned above as a means for storing all data. necessary for the implementation of the discrimination and attenuation processing method as described with reference to the figure 5 . This device receives as input successive frames of the digital signal Se and delivers the reconstructed signal Sa with pre-echo attenuation in the pre-echo areas discriminated with, if necessary, reconstruction of the attenuated signal by combining attenuated sub-signals.

The memory block BM may comprise a computer program comprising the code instructions for implementing the steps of the method according to the invention when these instructions are executed by a μP processor of the device and in particular the steps of calculating a control coefficient of the energies for at least two sub-blocks preceding the sub-block in which an attack is detected, comparing the steering coefficient to a predefined threshold and inhibiting the pre-echo attenuation processing in the pre-echo area in the case where the calculated coefficient of direction is lower than the predefined threshold.
The figure 5 can illustrate the algorithm of such a computer program.

This discrimination and attenuation processing device according to the invention can be independent or integrated in a digital signal decoder. Such a decoder can be integrated with equipment for storing or transmitting digital audio signals such as communication gateways, communication terminals or servers of a communication network.

Claims (11)

1. Method for discriminating and attenuating pre-echo in a digital audio signal generated from a transform coding, in which, upon decoding, for a current frame decomposed into sub-blocks, the low-energy sub-blocks preceding a sub-block in which a transition or onset is detected (E601) determine a pre-echo zone (E602) in which a pre-echo attenuation processing is carried out (E607), the method being characterized in that, in the case where an onset is detected from the third sub-block of the current frame, it comprises the following steps:

   - calculation (E603) of a leading coefficient of the energies for at least two sub-blocks of the current frame preceding the sub-block in which an onset is detected;

   - comparison (E604) of the leading coefficient to a predefined threshold; and

   - inhibition (E602) of the pre-echo attenuation processing in the pre-echo zone in the case where the calculated leading coefficient is below the predefined threshold.
2. Method according to Claim 1, characterized in that it further comprises a step of decomposition of the digital audio signal into at least two sub-signals as a function of a frequency criterion and in that the comparison calculation steps are performed for at least one of the sub-signals.
3. Method according to Claim 1, characterized in that it further comprises a step of decomposition of the digital audio signal into at least two sub-signals as a function of a frequency criterion and in that the computation and comparison steps are performed for each of the sub-signals, the inhibition of the pre-echo attenuation processing in the pre-echo zone of all the sub-signals being performed when a calculated leading coefficient is below the predefined threshold for at least one sub-signal.
4. Method according to Claim 3, characterized in that a different threshold is defined for each sub-signal.
5. Method according to one of Claims 1 to 4, characterized in that the leading coefficient is calculated according to a least squares estimation method.
6. Method according to one of Claims 1 to 5, characterized in that the leading coefficient is normalized.
7. Method according to Claim 1, characterized in that, in the case where an onset is detected in the first or second sub-block of the current frame, a leading coefficient calculated for the preceding frame is used for the comparison step.
8. Device for discriminating and attenuating pre-echo in a digital audio signal generated by a transform coder, the device being associated with a decoder and comprising a transition or onset detection module (601), a pre-echo zone discrimination module (602) and a pre-echo attenuation processing module (607), an echo attenuation processing being performed for a current frame decomposed into sub-blocks, in the low-energy sub-blocks preceding a sub-block in which a transition or onset is detected determining a pre-echo zone, the device being characterized in that it further comprises:

   - a computation module (603) calculating a leading coefficient of the energies for at least two sub-blocks of the current frame preceding the sub-block in which an onset is detected, in the case where an onset is detected from the third sub-block of the current frame;

   - a comparator (604) capable of performing a comparison of the leading coefficient to a predefined threshold; and

   - a discrimination module (602) capable of inhibiting the pre-echo attenuation processing in the pre-echo zone in the case where the calculated leading coefficient is below the predefined threshold.
9. Digital audio signal decoder comprising a pre-echo discrimination and attenuation device according to Claim 8.
10. Computer program comprising code instructions for implementing the steps of the method according to one of Claims 1 to 7, when these instructions are executed by a processor.
11. Storage medium that can be read by a pre-echo discrimination and attenuation processing device on which is stored a computer program comprising code instructions for executing the steps of the pre-echo discrimination and attenuation processing method according to one of Claims 1 to 7, when said storage medium runs on a computer.

Images