FR2911227A1 - Digital audio signal coding/decoding method for telecommunication application, involves applying short and window to code current frame, when event is detected at start of current frame and not detected in current frame, respectively - Google Patents

Abstract

The method involves providing two weighting windows of distinct lengths, and using a short window for coding a frame in which a specific event e.g. attack, has to be detected. The specific event in a current frame is detected for coding the current frame. The short window is applied in order to code the current frame, when the event is detected at the start of the current frame. A long window is applied to code the current frame, when the event is not detected in the current frame. Detecting and applying operations are repeated for a following frame. Independent claims are also included for the following: (1) an coder comprising a detection module (2) a computer program comprising a set of instructions to perform a method for coding/decoding of a digital audio signal (3) a decoder comprising a memory.

Description

Transform coding, using weighting and low windows

  The present invention relates to the encoding / decoding of digital audio signals.

  In a transform coding scheme, for a rate reduction, it is usually sought to reduce the precision given to the coding of the samples, ensuring that the listener can however perceive only the least amount of damage.

  To this end, the reduction in precision, performed by a quantization operation, is controlled from a psychoacoustic model. This model, based on a knowledge of the properties of the human ear, makes it possible to dose the quantization noise in the least perceptible auditory frequencies.

  In order to use the information from the psychoacoustic model, essentially given in the frequency domain, it is conventional to carry out a time / frequency transformation, the quantization taking place in this frequency domain.

  FIG. 1 schematically illustrates the structure of a transform coder, with: a bank BA of analysis filters FA1,..., FAn, attacking the input signal X, a quantization module Q, followed by a coding module COD, and a bank BS of synthesis filters FS1,..., FSn delivering the coded signal X '.

  To further reduce the rate of transmission, quantized frequency samples are encoded, often using so-called entropic coding (lossless coding). The quantization is performed in a conventional manner by a scalar quantizer, uniform or not, or by a vector quantizer.

  The noise introduced into the quantization step is shaped by the synthesis filter bank (also called inverse transformation). The inverse transformation, linked to the analysis transform, must therefore be chosen in order to focus the quantization noise, frequency or time, in order to avoid it becoming audible.

  The analysis transformation must best focus the signal energy to allow easy coding of samples in the transformed domain. In particular, the coding gain of the transform, which depends on the input signal, must be maximized as far as possible. For this purpose a relation of the type: SNR = GI • (+ K • R) is used where K is a constant term which can advantageously be 6.02 Thus, the signal-to-noise ratio obtained (SNR) is proportional to the number of bits per selected sample (R) plus the component Gu which represents the coding gain of the transform The higher the coding gain, the better the quality of reconstruction.

  We therefore understand the importance of transformation in coding. It ensures easy coding of samples, thanks to its ability to concentrate both the signal energy (by the analysis part) and the quantization noise (by the synthesis part).

  Since the audio signals are notoriously non-stationary, the time-frequency transformation must be adapted over time, depending on the nature of the audio signal.

  Some applications to conventional coding techniques are described below. In the case of modulated transforms, the standard audio coding techniques incorporate cosine modulated filter banks that enable these coding techniques to be implemented using fast algorithms, based on cosine transforms or transforms. Fast Fourier. Among the transformations of this type, the most commonly used transform (in MP3 encoding, MPEG-2 and MPEG-4 AAC in particular) is the Modified Discrete Cosine Transform (MDCT), the expression of which is shown below: 2M - I XI, '= xn + tMpk (n) 0k <M n = 0

  with the following notations: • M represents the size of the transform. • xn +, M are the samples of the digitized sound at a period 4 (inverse of the

  sampling frequency) at time n + tM, • t is the frame index. • Xk are the samples in the transformed domain for the t-frame,

  • pk (n) =, I M (n) cos [ù ((2n + 1-i-M) (2k + 1) 1 is a basic function of the transform whose h (n) is called prototype filter of size 2M.

  To restore the initial temporal samples, the following inverse transform is applied in order to reconstruct the samples 0 n M -1: M -1 1 xn + (M [xi. + 'Pk (n) + Xi, pk (n + M) 1 With reference to FIG. 1a, the reconstruction is carried out as follows: inverse DCT transform (hereinafter referred to as DCT-I) of the samples xk giving 2M samples, inverse DCT transform of the samples Xk giving 2M samples, the M first samples having a temporal support identical to the last M samples of the previous frame, k == 0 • weighting by the synthesis window h (M + n) for the second half of the frame T (last M samples), and by the synthesis window h (n) for the first half of the following frame T; +1 (M first samples), and • additions of the windowed components on the common support, in order to ensure the exact (so-called perfect) reconstruction of the signal. (depending on the condition xn +, shot = xn + tM), it is to choose a prototype window h (n) satisfying some constraints. Typically, the following relations are satisfactory in order to allow a perfect reconstruction: h (2Mu1ûn) = h (n) h2 (n) + h2 (n + M) = 1 the windows being evenly symmetric with respect to a central sample. It is relatively simple to satisfy these two simple constraints and, for this purpose, a conventional prototype filter consists of a sinusoidal window which is written as follows: h (n) = sin [_ (n + 0.5)] Well Of course, other prototype filter forms exist, such as windows defined in the MPEG-4 standard as Kaiser Bessel Derivatives (or KBD), or low overlay windows.

  An example of treatment with an MDCT transform, with long windows, is given in FIG. In this figure: the arrows in dashed lines illustrate a subtraction, the arrows in solid lines illustrate an addition, (1) the arrows in dashed mixed lines illustrate a DCT processing in the coding and DCT "decoding DEC, this term DCT corresponding to the cosine term of the basic function presented above, the samples of the signal to be coded are in a flux noted xin, and the evolution of the coding / decoding processes of certain samples surrounded and referenced a and b is followed. in FIG. 1b and e and f in FIG. 1c, the samples xin are grouped in frames, a current frame is denoted T, the previous and next frames being respectively denoted by Ti_1 and T; + i, - the reference DEC is proper to the processing carried out by the decoder (using synthesis windows FS with addition-overlap), the analysis windows are denoted FA and the synthesis windows are denoted FS, - n is the distance between the window medium and the sample a.

  The reference calc T '; is suitable for calculating the coded frame T '; using the analysis window FA and the respective samples T _1 and T frames. This is simply an example of a convention shown in Figure la. It could also be decided, for example, to index the frames T, and T; + A for the computation of a coded frame T '. Continuing with the example of FIG. 1a, the reference calc T '; + 1 is suitable for calculating the coded frame T'1 + 1 by using the respective samples of the frames Ti and T; +1.

  The terms v1 and v2 obtained before transformation DCT and inverse transformation DCT- 'are obtained with equations of the type: v1 = a * h (M + n) + b * h (2 * M-1-n), and v2 = b * h (M-1-n) - a * h (n) Thus, after overall treatment DCT / DCT '' and window of synthesis, the terms of reconstruction a 'and b' are written: a '= vl * h (M + n) - v2 * h (n) = a * h (M + n) * h (M + n) + b * h (2 * M-1-n) * h (M + n) - b * h (M-1-n) * h (n) + a * h (n) * h (n), and b '= v1 * h (2 * M-1-n) + v2 * h (M) -1-n) a * h (M + n) * h (2M-n1) + b * h (2 * M- • 1-n-1) * h (2M-n-1) + b * h ( M-1-n) * h (M-1-n) a * h (n) * h (M-1-n) and we verify that the reconstruction is perfect (a '= a and b' = b). (using relations (1) and deducing h (M-1-n) = h (n + M)) The principle described above of an MDCT transform naturally extends to transformations named ELT (Extended Lapped Transform), in which the order of the basic functions is greater than twice the size of the transform, with in particular:

  Xk = E xn +, M pk (n), where 0 k <M and L = 2KM, where K is a positive integer greater than 2. n = 0 For reconstruction, instead of associating two consecutive frames as for an MDCT transform , the synthesis of the samples implements K successive frames windowed.

  It is further indicated that the window symmetry constraint (principle described in detail below) may be relaxed for an ELT-type transform. The identity constraint between the analysis and synthesis windows can also be relaxed and this is called bi-orthogonal filters.

  Given the need to adapt the transform to the signal to be encoded, the prior art allows what is called a window switching, i.e. a change over time of the size of the transform used. The need to change window length can be justified in particular in the following case. When the signal to be encoded, for example a speech signal, comprises a transient signal (non-stationary) characterizing a strong attack (for example the pronunciation of a sound ta or pa characterizing a plosive in the speech signal), it is necessary to increase the temporal resolution of the coding and hence reduce the size of the coding windows, which therefore requires going from a long window to a short window. Very precisely, in the known prior art, in this case a long window (FIG. 2a, which will be described later) is passed to a transition window (FIG. 2c described later), then to a series of short windows ( Figure 2b described later). It is therefore necessary to anticipate an attack on at least one next frame, as will be seen in detail below, before deciding the length of the coding window of a current frame, and therefore, code the current frame.

  An example of a change in window length in the sense of the prior art is described below.

  A typical example is the size change from an MDCT transform of size m to a size m / 8, as provided in the MPEG-AAC standard.

  In order to preserve the property of perfect reconstruction, equation (1) above must be replaced by the following formulations at the time of the transition between two sizes: J h2 (n) + h2 (Mnn) = 1 for OSn < M th2 (M + n) + h2 (2Mnn) = 1 otherwise A relation is given in addition for consecutive prototype filters of different sizes: 20 hI (M + M / 2μMs / 2 + n) = h2 (Ms.ùn) 0 <ùn <M4 There is a symmetry around the size M / 2 at the moment of the transition.

  Different types of window have been illustrated in FIGS. 2a to 2e, with respectively: a sinusoidal window (symmetrical sinus function) of size 2M = 512 samples for FIG. 2a, a sinusoidal window (symmetrical sinus function) of size 2Ms = 64 samples for Figure 2b, a transition window to go from a size 512 to 64 for Figure 2c, a transition window to go from a size of 64 to 512 for Figure 2d, and an example of construction done using the basic windows presented above, for figure 2e. Each succession is of predetermined length defining what is called the window length. Thus, samples to be encoded are combined, at least two by two, and weighted, in combination, by respective values of weighting of the window, as seen with reference to FIG. More particularly, the sinusoidal windows (FIGS. 2a and 2b) are symmetrical, that is to say that the weighting values are substantially equal on both sides of a central value, in the middle of the succession of values. forming the window. An advantageous embodiment consists in choosing sinus functions to define the variations of weighting values of these windows. The other window choices remaining possible (for example those used in MPEG AAC coders).

  On the contrary, it should be noted that the transition windows (FIGS. 2c and 2d) are asymmetrical and comprise a flat region (reference PLA), which means that the weighting values in these regions are maximum and for example are equal to 1. As will be seen with reference to Figs. 1b and 1c, using a transition window from a long window to a short window (Fig. 2c), two samples (in the example shown in Fig. 1b), of which sample a, are simply weighted by a factor 1, while the sample b is weighted by the factor 0 in the calculation of the coded frame T ', so that the two samples whose sample a are simply transcribed such which (the DCT excepted) in the coded frame T ';.

  The use of a variable size transform in a coding system is described below. Operations at the level of a decoder which must reconstruct the audio samples are also described. In standard systems, the coder most often selects the transform to be used over time. Thus, in the AAC standard, the encoder transmits two bits for selecting one of the four window size configurations presented above.

  The MDCT processing using transition windows (long-short) is illustrated in FIGS. 1b and 1c. These figures represent the calculations carried out, in the same way as for the figure la.

  In FIGS. 1b and 1c, only a few short analysis windows, referenced FA (with Ms = M / 2 in the illustrated example) are shown. Nevertheless, in reality, as illustrated in FIG. 2e, a succession of several short windows (with typically Ms = M / 8) is provided. It will therefore be understood that each window FA of FIGS. 1b and 1c actually includes a succession of short windows.

  The FTA transition window (FIG. 1b), for calculating the coded frame T '; (reference shim T ';) comprises: - a half-window long on M: samples, in its rising edge, and, in its falling edge: o a first flat region PLA (with weighting values equal to 1) on ( M / 2-Ms / 2) samples, o a half-short window down on Ms samples, and o a second flat region (with weighting values equal to 0) on (M / 2-Ms / 2) samples.

  For the calculation of the following coded frame T '; + 1 (reference calc T'; + i), the first (M / 2-Ms / 2) samples are ignored and therefore not processed by the short windows, the following Ms samples are weighted by the rising edge of the short analysis window FA as shown in FIGS. 1b and 1c, and the following Ms samples are weighted by its falling edge. In the following, the following notations are used: - M is the long frame size, - Ms is the short frame size.

  In Figure lb, sample b is synthesized using only the short windows to respect the analogy with the calculation for long windows. Then, because of the particular shape of the long-short transition half-window, sample a is reconstructed directly from the analysis and synthesis transition windows.

  The transition window is referenced FTA in FIGS. 1b and 1c.

  In Figure 1c, the samples corresponding to the transition zone between the long-short window and the short window are calculated. By analogy with the calculation for the long windows of Figure la, here we follow the processing of samples noted e and f (surrounded).

  Two examples of window transition situations are described below.

  In a first example, an attack requiring the use of short windows in the audio signal at a time t = 720 (FIG. 2e) has been detected. The encoder must indicate to the decoder the use of long-short transition windows coming between the long windows previously used and the short, subsequent windows.

  Thus, the encoder indicates to the decoder, successively, the sequences: • long window • long-short transition window • short window • short-long transition window • long window.

  The decoder then applies a relation of the type:

  M-1 1 xn + rM ù [x 'p'k (n) + XF p`k (n + M)] k = 0 where pk and pk represent the synthesis functions of the transforms at times t and t + 1, which can be distinct from each other. The reconstruction is done as before, with the difference that if the basic functions pk and pi 'are of different sizes, then, with reference to FIG. 1b, an inverse DCT transform of size M of samples Xk giving 2M 10 samples, • an inverse DCT transform of size Ms of samples Xk 'giving 2Ms samples, the Ms first samples having a common temporal support of length Ms in a recovery zone comprising the rising part of the short window, with the samples coming from of the inverse transform DCT of size M of the falling part of the transition window FTA, a multiplication by the dual synthesis window of the transition window FTA and referenced FTS in FIG. 1b, for the first half, and a multiplication by the short synthesis window FS for the second half, and • the additions of these windowed components on the recovery area Thus, the time support corresponding to an end portion of the initial frame T. The decoder is therefore slave to the encoder and faithfully applies the window types decided by the encoder.

  In this first example, the encoder detects a transition upon the arrival of the samples of a first frame (for example the frame 1 of FIG. 2e, comprising the samples between the instants t = 512 and t = 767). The encoder can then decide that the current window should be a long-short transition window, encoded, transmitted and signaled to the decoder. Then, eight short windows are successively applied between the samples t = 624 and t = 911. Thus, at the time of the transition (t = 720), the encoder uses the short windows, which allows a better temporal representation of the signal.

  In a second example, a transition is detected at sample t = 540. When the encoder receives the samples of a first frame (the frame 0 of Figure 2e for example), it does not detect a transition and therefore selects a long window. At the arrival of the samples of a second frame (frame 1 in the example of Figure 2e), next, the encoder detects an attack (at time t = 540). In this case then, the detection is performed too late and the use of a transition window does not benefit from the use of short temporal supports (short windows) at the time of the attack. The encoder must then anticipate the use of short windows and, therefore, insert an additional coding delay corresponding to at least M / 2 samples.

  It will then be understood that a disadvantage of the known prior art lies in the fact that it is necessary to introduce an additional delay to the encoder in order to detect an attack in the time signal of a next frame and to then anticipate the passage to short windows. This attack may correspond to a transient signal of high intensity, such as a plosive, for example, in a speech signal, or to the appearance of a percussive signal in a musical sequence.

  In some telecommunications applications, the additional delay necessary for the detection of transient signals and the use of transition windows is not acceptable. For example, in the MPEG-4 AAC Low Delay encoder, short windows are not used, only long windows are allowed.

  The present invention improves the situation.

  It proposes a transition between windows not requiring the introduction of an additional delay.

  To this end, it aims at a coding / decoding method, by transform, of a digital audio signal represented by a succession of frames, in which: at least two weighting windows of distinct respective lengths are provided, and a short window is used. to encode a frame in which a particular event has been detected. This particular event may be for example a non-stationary phenomenon, such as a strong attack present in the digital audio signal that contains the current frame.

  More particularly, for the coding of a current frame, it is sought to detect the particular event in this current frame, and: at least if the particular event is detected at the beginning of the current frame, a short window is applied to code the current frame, - while if the particular event is not detected in the current frame, a long window is applied to code the current frame. These steps are repeated for a next frame, so that it is possible, within the meaning of the invention, to code a given frame using a long window and to code a frame that immediately follows this given frame by directly using a short window, without using a transition window as in the prior art.

  By offering the possibility of going directly from a long window to a short window, the detection of the particular event can be conducted directly on the frame being encoded and no longer on the following frame, as in the art. prior. Therefore, a coding carried out by the method in the sense of the invention is carried out without additional delay with respect to a fixed size MDCT transform, contrary to the encodings of the prior art.

  Other features and advantages of the invention will appear on examining the detailed description below, and the appended drawings in which, besides FIGS. 1, 1a, 1b, 1c, 2a, 2b, 2c, 2d, 2, relating to the prior art and described above: FIG. 3a schematically illustrates a coding / decoding process in the sense of the invention, following the evolution of samples a and b, as in FIG. 1b previously described. FIG. 3b diagrammatically illustrates a coding / decoding process in the sense of the invention by following the evolution of samples e and f, as in the figure described previously, and FIGS. 4a and 4b illustrate examples of variation. of the weighting functions used for the decoding compensation carried out in the implementation of the invention, FIG. 5a illustrates an example of a process which can be applied in an encoder within the meaning of the invention, FIG. example of treatment that It can be applied in a decoder in the sense of the invention, and FIG. 6 illustrates the respective structures of an encoder and a decoder and the communication of the information of the type of window used to the encoding.

  The present invention avoids applying transition windows at least for the passage of a long window to a short window. Thus, by taking up the second example described above with reference to FIG. 2e, if a non-stationary phenomenon or attack is detected at time t = 540, the present invention proposes to use a long window for frame 0 (window extending from time t = 256 to time t = 511). Then, when acquiring the samples from the next frame (t = 512 at t = 767) and detecting an attack at t = 540, the encoder uses eight short windows to encode the samples from the sample. instant t = 368 (corresponding to t = 512-M / 2-Ms / 2), up to the instant t = 655 (corresponding to t = 5 12 + M / 2 + Ms / 2- 1, where - 2 * M = 512 is the size of the long window, and - 2 * Ms = 64 is the size of the short window, in the example described), and this, without a conventional transition window, asymmetric like that represented on the Figures lb and 1c relating to the prior art.

  At the decoder, when receiving the encoded frame with short windows, the decoder then proceeds to the following operations: • reception of information from the coder indicating that short windows must be used for the current frame, • application advantageous to compensate for the direct transition from a long window to a short window to the coding, an example of this processing being described in detail below, with reference to FIG. 5b. FIGS. 3a and 3b illustrate the coding / decoding method within the meaning of the invention to obtain on the one hand the samples a and b which are in an area without overlap between the long and short windows (FIG. 3a), and on the other hand the samples e and f being in this overlap zone (FIG. 3b). In particular, this overlap area is defined by the falling edge of the long window FL and the rising edge of the first short window FC.

  Thus, with reference to FIGS. 3a and 3b, to the coding, the samples of the frames T; _1 and T; are weighted by the long analysis window FL to constitute the coded frame T '; And the samples of the following frame T; and T; + i are weighted directly by short analysis windows FC, without applying a transition window.

  It will also be noted, with reference to FIGS. 3a and 3b, that the first short analysis window FC is preceded by values not taken into account by the short windows (for the samples preceding the sample e in the example of FIG. Figure 3b). More particularly, this processing applies to the first M / 2-Ms / 2 samples of the frame to be coded T'1 + 1, and this, similarly to the coders / decoders of the prior art. In general, the aim is to disturb as little as possible the processing carried out on coding, as on decoding, compared to the prior art. Thus, for example, it is chosen to ignore the first samples of the coded frame T'1 + 1. Of course, in FIGS. 3a and 3b, only the case of two short windows (Ms = M / 2) for FC analysis is shown. However, as in the prior art, there is provided a succession of several short windows and each succession of short windows is illustrated in these figures 3a and 3b bearing the reference FC.

  In what follows, two exemplary embodiments are described for decoding a frame T'1 + 1 which has been coded using a short window FC while an immediately preceding frame T'1 has been coded using a long window FL. In a first embodiment, it completely eliminates the use of synthetic windows decoding and it is shown that the property of perfect reconstruction is ensured.

  In FIG. 3a, when detecting an attack requiring a window change (from a long window directly to a short window), the samples are first synthesized from the short windows only (sample b on the window). Figure 3a). Then, in the value v1 calculated from the long analysis window, the effect of the sample b calculated in advance is compensated. The coding calculation (coded frame T ') for the sample then takes place as follows: v1 = a * h (M + n) + b * h (2 * M-1-n). On the other hand, the sample a is not weighted in the coding value v2 since the calculation of weighting from the short window followed by a combination is done on a separate temporal support (coded frame T'1 + 1), and we have after reconstruction from the short windows: 15 v2 = b

  The perfect reconstruction in the coding / decoding in the sense of the invention is advantageously verified. Indeed: a '= (vi - v2 * h (2 * M-1-n)) / h (M + n) = a It will also be noticed, at the decoding, that the samples drawn from the values v2 = b and the following ones must be determined first, before determining the samples at the beginning of the frame (as sample a). This results in a time reversal during decoding.

  In FIG. 3b, the coded samples of the transition zone between the long window FL (falling edge) and the first short window FC (rising edge) are calculated, thus at the level of the samples e and f. The expression of the coded coefficients (or values v1 and v2 hereinafter) is given, in this overlap zone between the two windows FL and FC, by the following equations: v1 = e * h (M + n) + f * h (2 * M-1-n), and v2 = f * hs (Ms-1-m) - e * hs (m) At the decoder, we must solve this system of equations with two unknowns to find the values samples e and f: e = [v1 * hs (Ms-1-m) -v2 * h (2 * M-1-n)] / [h (M + n) * hs (Ms-1-m) + hs (m) * h (2 * M-1-n)] f = [v1 * hs (m) + v2 * h (M + n)] / [hs (Ms-1-m) * h (M + n) + h (2 * M-1-n) * hs (m)] We deduce again the formulas satisfying the property of perfect reconstruction: e '= [vi * hs (Ms + m) - v2 * h (n)] / [h (M + n) * hs (Ms + m) + h (2 * M-1-n) * hs (m)] = e, and f = [vl * hs (2 * Ms -1-m) + v2 * h (M-1-n)] / [h (M + n) * hs (Ms + m) + h (2 * M-1-n) * hs (m)] = f, with m = n ù M / 2 + Ms / 2 1730 It will be noted that the value v2 is weighted by the long window h, and this, contrary to what was established in the prior art (where v2 was weighted p ar the short window hs as shown at the bottom of Figure 1c).

  In a second embodiment, synthetic windows are preserved during decoding. They are of the same form as the analysis windows (homologous or dual analysis windows), as illustrated in FIGS. 3a and 3b and bearing the reference FLS for a long, synthetic window, and the FCS reference for a short, synthetic window. This second embodiment has the advantage of being in accordance with the operation of the decoders of the state of the art, namely to use a long synthesis window to decode a frame which has been coded with a long analysis window and use a series of short summary windows to decode a frame that has been encoded with a series of short analysis windows.

  On the other hand, a correction of these synthesis windows, by compensation, is applied to decode a frame which has been coded with a long window, whereas it should have been coded with a long-short transition window. In other words, to compensate for the effect of the direct passage from a long window to a short window, to the encoder, the following processing is performed to decode a current frame T '; + 1 which has been coded using a short window FC while an immediately preceding frame T '; had been coded using a long FL window.

  The above equations for decoding and binding samples a, b, e, f to values v1 and v2, can be rewritten as weighted sums of two terms, as follows, in particular by time reversal. .

  First, one places oneself in the first synthetic FCS short windows and after the aforementioned overlap zone (typically to the sample v2 = b and the following ones in the illustrative illustration of FIG. 3a). For the decoding of this non-overlapping part, from short synthesis windows FCS only, the values of the coded frame T '; + 1 are first decoded from v2 = b (FIG. 3a). Once

  that the samples b and the following ones are decoded, one applies the weighted sum of two terms following: xn = w1nln + w2, nsm_1-n, with 0 <-n <M / 2-Ms / 2, where: xn represents the samples decoded (corresponding to the initial samples xn, since the coding / decoding is perfect reconstruction), the notation ln designates what would correspond to samples that would have been decoded (application of a DCT inverse transformation "1) using a synthesis window long FLS, without correction, and - s ,, represents the completely decoded samples (typically the sample b and the samples which follow it) by means of the succession of short FCS synthesis windows The two weighting functions wl, , and w2, then write:

  Won = 1 and w2n - h (2Mu11) - h (n), with 0 <ùn <M / 2μMs / 2h- (M + n) h (M + n) h (M + n) It will be understood that the -in samples are actually incompletely decoded values by synthesis and weighting using the long synthetic window. These are typically the values v1 of FIG. 3a, multiplied by the coefficients h (M + n) of the window FLS, and in which still occur samples of beginning of the frame T1, such as the sample a. It will also be noted that the samples b and the following are here determined first and are written in the formula given above SM _, _ n, thus illustrating the time reversal that the decoding process proposes in this second embodiment.

  We also note that the weighting performed by the long FLS synthesis window is avoided because it is removed from the term wi ,,, (because of the division by h (M + n)).

  Moreover, for the reconstruction of the portion of samples covered by both the long window FL (falling edge) and the first short window FC (rising edge), corresponding to the region of the samples e to f of FIG. 3b, rather apply the following combination of two weighted terms: xn = WinSm + W2nln, with m = nùM / 2 + Ms / 2 and M / 2ùMs / 2 <_n <M / 2 + Ms / 2 As previously, the terms ln constitute the incompletely reconstituted values by synthesis and weighting by the FLS long synthesis window and the sn terms represent the incompletely reconstituted values from the rising edge of the first short synthesis window FCS.

  The weighting functions w'1 n and w'2 n are given here by: h (n) ùh, (~) h (Ms -1μm) h, (Ms-1ùm) h (Mùnn) h (Mùnn) 7ù7, w hn h (Mi1n) h, (M, -1-m) + h (n) h, (m) and w2n = h (M-1n) h, (M, -1m) + h (n) h (m) All these weighting functions w1 ,, wz n, w, '", and wz n consist of fixed elements that depend only on long and short windows, and examples of variations of such weighting functions are presented. In an advantageous embodiment, the values taken by these functions can be calculated a priori (tabulated) and stored permanently in memory of a decoder within the meaning of the invention.

  Thus, with reference to FIG. 5b, the decoding processing of a frame T '; which has been coded by passing directly from a long analysis window to a short analysis window, can comprise the following steps, in an exemplary embodiment. To decode the frame T '; (step 60), a short synthesis window (step 61) is first applied to decode the end-of-frame value v2 == b (step 63). Here we rely on a following coded frame T '; + 1 (step 62) to determine b. A long synthesis window (step 64) is then applied to decode the start-of-frame samples T '; (step 65), applying the compensation for all n between 0 and M / 2-Ms / 2 using the relation xn = wi nn + w, nsM_l_n (step 67) and using the weighting values pre-calculated and tabulated w,, and w2 n (step 66). Decoding the central region of the coded frame T '; (between e and f), so for n between M / 2-Ms / 2 and M / 2 + Ms / 2, can be carried out in parallel (+ sign of Figure 5b) using both the synthesis windows short and long (step 68) and applying in particular the compensation (step 69) from the relation xn = WI.nS m + W2, nln, where m = n - M / 2 + Ms / 2 and with the values weighting and W'2, n pre-calculated and tabulated (step 70). Finally, from this processing (step 71), the values of all the sample types a, b, e or f of the initial frame T. are deduced.

  The first and second embodiments described above, the decoding of a frame T '; which has been coded by passing directly from a long analysis window to a short analysis window, guarantee a perfect reconstruction and then allow the coding to effectively switch from a long window to a short window, directly.

  An exemplary embodiment in which it is proposed to dispense with the coding of the application of a long-short transition window, at least in certain cases, will now be described with reference to FIG. 5a.

  Upon reception of a frame Ti (step 50), the presence of a non-stationary phenomenon, such as an ATT attack (test 51), is sought in the digital audio signal present directly in this frame Ti. As long as no such phenomenon is detected (arrow n at the output of the test 51), the application of long windows (step 52) is continued for the coding of this frame Ti (step 56). Otherwise (arrow o at the output of the test 51), it is sought to determine whether the ATT phenomenon is at the beginning (for example in the first half) of the current frame Ti (test 53), in which case (arrow o at the output of the test 53 ) a short window (step 54), and more exactly a series of short windows, is directly applied for the encoding of the Ti frame (step 56). This embodiment then makes it possible to avoid a transition window and not to wait for the next frame T1 + 1 to apply a short window.

  Thus, it will be understood that, unlike the state of the art, a particular event such as a non-stationary phenomenon can be detected directly in the frame being encoded Ti and not in a following frame T; +1. The coding delay in the sense of the invention is then reduced compared to that of the prior art. Indeed, if the non-stationary phenomenon is detected at the beginning of the current frame, a short window is directly applied, whereas in the prior art, it would have been necessary to detect the non-stationary phenomenon in a following frame T; +1 so as to can apply a transition window to the frame being encoded Ti.

  Referring again to FIG. 5a, if the non-stationary phenomenon is detected at the end (for example in the second half) of the current frame Ti (arrow n at the output of the test 53), it is advantageous to choose to apply a window of transition (step 55) for coding the current frame Ti (step 56), before applying a succession of short windows. This embodiment makes it possible in particular to propose a treatment equivalent to that of the state of the art, while ensuring a reduced coding delay.

  Thus, in more generic terms, at least three weighting windows are provided in this embodiment: a short window, a long window, and a transition window to switch from using the long window to using the window. short, and if a particular event such as a non-stationary phenomenon is detected at the end of the current frame (step 53), applying a transition window (step 55) to code (step 56) the current frame (T;).

  In a variant of this embodiment, it can be provided, for a transition between a long window use to a short window use: for a current frame Ti, to use a long window FL, and for an immediately consecutive frame T; +1, to directly use a short window FC, without using a transition window, even though the particular event would be detected at the end of the current frame.

  This variant has the following advantage. Since the encoder must send to the decoder a window type change information, this information can be coded on a single bit because there is no more to inform the decoder of the choice between a short window and a transition window.

  Nevertheless, it is possible to maintain a transition window to go from a short window to a long window and, in particular, to still transmit the one-bit window type change information after receiving information. the decoder can for this purpose: use the short window, - then, in the absence of reception of window type change information, use a transition window of a window. short window to a long window, - then finally, use a long window.

  FIG. 6 illustrates the information communication of the type of window used for encoding, from an encoder 10 to a decoder 20. It will be recalled that the encoder 10 comprises a detection module 11 for a particular event such as a strong attack in the signal contained in a frame T; during encoding and that it deduces from this detection the type of window to use. For this purpose, a module 12 selects the type of window to be used and transmits this information to the coding module 13 which delivers the frame T '; coded using the analysis window FA selected by the module 12. The coded frame T '; is transmitted to the decoder 20, with the information INF on the type of window used in the coding (generally in the same stream). The decoder 20 comprises a synthesis window selection module 22 according to the information INF received from the encoder 10 and the module 23 applies the decoding of the frame T '; The present invention also aims at an encoder such as the encoder 10 of FIG. 6 for carrying out the method within the meaning of the invention and more particularly for carrying out the treatment illustrated in FIG. 5a. or its previously described variant (transmitting the window type change information on a single bit).

  The present invention also relates to a computer program, intended to be stored in memory of such an encoder and comprising instructions for the implementation of such a treatment, or its variant, when such a program is executed by a processor of the encoder. As such, Figure 5a may represent the flowchart of such a program.

  It will be recalled that the encoder 10 uses analysis windows FA and the decoder 20 can use synthesis windows FS, according to the second embodiment mentioned above, these synthesis windows being homologous to the analysis windows FA, while nevertheless proceeding to the compensation correction described above (using the weighting functions wi ,,,, w2 ,,, wet w '2, n).

  The present invention also aims at another computer program, intended to be stored in memory of a transform decoder such as the decoder 20 illustrated in FIG. 6, and comprising instructions for implementing the decoding according to the first embodiment. , or according to the second embodiment described hereinabove with reference to FIG. 5b, when such a program is executed by a processor of this decoder 20. As such, FIG. 5b can represent the flowchart of such a program.

  The present invention also relates to the transform decoder, itself, then comprising a memory storing the instructions of a computer program for decoding.

  In generic terms, the process of decoding, in the context of the invention, by transforming, a signal represented by a succession of frames which have been coded using at least two types of weighting windows, of respective respective lengths, is as following.

  In case of reception of information passing from a long window to a short window: samples (of type b) are determined from a decoding applying a type of short synthesis window FCS to a given frame T ' +1 which has been coded using a short analysis window FC, and complementary samples are obtained by: • partially decoding (application of a DCT-1 inverse transform) a frame T'1 preceding the given frame and which was encoded using a long FL analysis window type, and • applying a combination of two weighted terms involving weighting functions that can be tabulated and stored in a decoder memory.

  In the second embodiment above, these are the functions denoted w1 ,,,, W2, n, w'l, n, W'2, n.

  However, this generic decoding treatment applies in both cases to the first and second embodiments. In the second embodiment: the samples (b) obtained from (63) are determined first (step 63 of FIG. of the given frame (T '; + 1), and samples (a) corresponding temporally at the beginning of the previous frame (T';) are deduced (steps 65-67), these samples being derived from a decoding applying a long FLS synthesis window and specific to the second embodiment.

  In this case, for: a frame comprising M samples, a long window comprising 2M samples, a short window comprising 2Ms samples, Ms being less than m, the samples xn, for n between 0 and (M / 2-Ms / 2 ), n = 0 corresponding to the beginning of a frame being decoded, are given by a combination of two weighted terms of the type: xn = W1 nln + W2 nSMu1nn, Oë: - in are values (v1) from the previous frame T ';, sM_1_n are samples (b) already decoded using short synthesis windows applied to the given frame and 20 - w,, and w2, are weighting functions whose values taken as a function of n can be tabulated and stored in the decoder memory.

  Otherwise, for n between (M / 2-Ms / 2) and (M / 2 + Ms / 2), the samples Xn are given by a combination of two weighted terms of the type: 25 xn ù WI, nS m + W2 , nln, where m = n ù M / 2 + Ms / 2, or: - in are values v1 from the previous frame T ';, - S m are values v2 from the given frame T'; + 1 , and - w'1 n and W'2, n are weighting functions whose values taken as a function of n can be as tabulated and stored in the decoder memory.

  The present invention therefore makes it possible to offer the transition between windows with a reduced delay compared with the prior art while retaining the property of perfect reconstruction of the transform. This method can be applied with all types of windows (unsymmetrical windows and windows of analysis and synthesis different) and for different transforms and banks of filters.

  The compensation treatments presented above in the case of a transition from a long window to a shorter window extend naturally and analogously to the case of a transition from a short window to a window of larger size. In this case, the absence of a short-long transition window can be compensated for by the decoder by a weighting analogous to the case presented above.

  The invention then applies to any transform coder, in particular those intended for interactive conversational applications, as in the MPEG-4 AAC-Low Delay standard, but also to different transforms of the MDCT transforms, in particular the abovementioned transforms ELT (FIG. for Extended Lapped Transform) and their bi-orthogonal extensions.

Claims (14)

  A method of encoding / decoding, by transforming, a digital audio signal represented by a succession of frames, wherein: providing at least two weighting windows, of respective distinct lengths, and using a short window to encode a frame in which a particular event has been detected, characterized in that, for coding a current frame (Ti), it is sought to detect (51) the particular event in said current frame (Ti), and: - at least if the particular event is detected at the beginning of the current frame (53), applying a short window (54) to code (56) the current frame (Ti), - if the particular event is not detected in the frame current, a long window (54) is applied to encode (56) the current frame (Ti), and these steps are repeated for a next frame (T, + 1).   2. Method according to claim 1, characterized in that said particular event is a non-stationary phenomenon.   3. Method according to one of claims 1 and 2, wherein there are provided at least three weighting windows: a short window, - a long window, and a transition window to switch from a use of the long window to a use of the short window, characterized in that if the particular event is detected at the end of the current frame (53), the transition window (55) is applied to encode (56) the current frame (Ti).   4. Method according to one of claims 1 and 2, characterized in that, for a transition between a long window use to a short window use: 28 for a current frame, the long window is used, and for a frame immediately consecutively, the short window is used directly, without using a transition window, and in that window type change information, sent from an encoder to a peer decoder, is encoded on a single bit.   5. Method according to one of the preceding claims, characterized in that a decoder receiving a passage information from the long window to the short window: uses the short window, - then, in the absence of receipt of information from change of window type, use a transition window from a short window to a long window, - then use a long window.   6. Method according to one of the preceding claims, wherein a decoder uses synthesis windows, dual analysis windows that uses a coder homologous decoder, characterized in that decoding a frame (T ' encoded by passing from a long window (FL) to a short window (FC), compensation is conducted and the coding / decoding has a property of perfect reconstruction of the transform.   A process for decoding, by transforming, a signal represented by a succession of frames that have been encoded using at least two types of weighting windows, of respective respective lengths, in which, upon receipt of information passing from a long window to a short window: - samples (b) are determined (63) from a decoding applying a short synthesis window type (61) to a given frame (T '; + 1) ) which has been coded using a short analysis window, and - complementary samples (67, 69) are obtained by: • partially decoding (DCT-I) a frame (T ';) preceding the given frame and which has was coded using a long analysis window type, and • applying a combination of two weighted terms involving weighting functions (w 1, n, w 2, n, w '1, n, w 2, ,) tabulated and stored in memory of a decoder.   8. The method as claimed in claim 7, characterized in that: (63) samples (b) originating from the given frame (T'1 + 1) are first determined, and samples (65-67) are deduced therefrom. (a) temporally corresponding to the beginning of the previous frame (T'1), these samples being derived from a decoding applying a long synthesis window.   9. Method according to claim 8, in which: a frame comprises M samples, a long window comprises 2M samples, a short window comprises 2Ms samples, Ms being less than m, characterized in that the samples xn, for n between 0 and (M / 2-Ms / 2), n = 0 corresponding to the beginning of a frame being decoded, are given by a combination of two weighted terms of the type: = = WI nln + w2 nSMu1nn, Oë: - in are values (v1) from the previous frame (T'1), are samples (b) already decoded using short synthetic windows applied to the given frame (T'1 + 1), and - w, n and w2 n are weighting functions whose values taken as a function of n are tabulated and stored in the decoder memory.   10. Method according to one of claims 7 to 9, wherein: - a frame comprises M samples, - a long window comprises 2M samples, - a short window has 2Ms samples, Ms being less than m, characterized in that the samples, for n between (M / 2-Ms / 2) and (M / 2 + Ms / 2), n = 0 corresponding to the beginning of a frame being decoded, are given by a combination of two terms weighted of the type: xn = wl, ns, n + w2, nln, where m = nM / 2 + Ms / 2, where: - lä are values (v1) from the previous frame (T ';), - S -m are values (v2) from the given frame (T '; + 1), and - w', n and w'2 ä are weighting functions whose values taken as a function of n are tabulated and stored in decoder memory.   11. Transform encoder, of a digital audio signal represented by a succession of frames, the encoder using at least two weighting windows, of respective respective lengths, characterized in that it comprises at least: a detection module (1l) of a particular event in a current frame to be coded (Ti), a selection module (12): * a short analysis window for coding the current frame (Ti) at least if the particular event is detected in beginning of the current frame, and * a long analysis window for coding the current frame (Ti) if the particular event is not detected in the current frame, and a coding module (13), applying the window analysis method selected by the selection module (12) for encoding the current frame (Ti).   12. Computer program, intended to be stored in memory of an encoder according to claim 11, characterized in that it comprises instructions for carrying out the method according to one of claims 1 to 6, when it is executed by a processor of such an encoder.   13. Computer program intended to be stored in memory of a decoder by transform, characterized in that it comprises instructions for implementing the decoding method according to one of claims 7 to 10, when it is executed. by a processor of such a decoder.   14. Transform decoder, characterized in that it comprises a memory storing the instructions of a computer program according to claim 13.