EP2452336A1 - Improved coding /decoding of digital audio signals - Google Patents

Abstract

The invention pertains to a method of hierarchical coding of a digital audio frequency input signal into several frequency sub-bands comprising a core coding of the input signal according to a first throughput and at least one enhancement coding of higher throughput, of a residual signal, the core coding using a binary allocation (506) according to an energy criterion. The method is such that it comprises the following steps for the enhancement coding: - calculation of a frequency-based masking threshold (511) for at least part of the frequency bands processed by the enhancement coding; - determination (512) of a perceptual importance per frequency sub-band as a function of the masking threshold calculated and as a function of the number of bits allocated for the core coding; - binary allocation (512) of bits in the frequency sub-bands processed by the enhancement coding, as a function of the perceptual importance determined; and – coding of the residual signal (513) according to the bit allocation. The invention also pertains to a suitable method of decoding, a coder and a decoder.

Description

 Advanced coding / decoding of digital audio signals

The present invention relates to a sound data processing.

 This processing is adapted in particular to the transmission and / or storage of digital signals such as audio-frequency signals (speech, music, or other).

 The invention applies more particularly to hierarchical coding (or "scalable" coding) which generates a so-called "hierarchical" bitstream because it comprises a core rate and one or more improvement layer (s). The G.722 standard at 48, 56 and 64 kbit / s is an example of a scalable bit rate codec, while the ITU-T G .129.1 and MPEG-4 CELP codecs are examples of scalable codecs in both bitrate and bitrate. in bandwidth.

 Hierarchical coding, having the capacity to provide varied bit rates, is described below by distributing the information relating to an audio signal to be coded in hierarchical subsets, so that this information can be used in order of importance. in terms of audio rendering quality. The criterion taken into account for determining the order is a criterion for optimizing (or rather reducing) the quality of the coded audio signal. Hierarchical coding is particularly suited to transmission over heterogeneous networks or having variable available rates over time, or to transmission to terminals with varying capacities.

 The basic concept of hierarchical audio coding (or "scalable") can be described as follows.

 The bit stream includes a base layer and one or more enhancement layers. The base layer is generated by a fixed rate codec, referred to as

"Heart coded", guaranteeing the minimum quality of the coding. This layer must be received by the decoder to maintain an acceptable level of quality. Improvement layers are used to improve quality. However, they may not all be received by the decoder.

 The main advantage of hierarchical coding is that it allows an adaptation of the bit rate simply by "truncation of the bit stream". The number of layers (i.e., the number of possible truncations of the bitstream) defines the granularity of the coding. We speak of coding with "high granularity" if the bit stream comprises few layers (of the order of 2 to 4) and coding with "fine granularity" allows for example a step of the order of 1 to 2 kbit / s.

 More specifically, the scalable bandwidth and scalability encoding techniques are described below, with a CELP core-type coder, in a telephone band, and one or more broadband enhancement layer (s). An example of such systems is given in the ITU-T G.729.1 8-32 kbit / s fine grain standard. The G.729.1 coding / decoding algorithm is summarized below.

* Reminders on the G.729.1 encoder

The G.729.1 encoder is an extension of the ITU-T G.729 coder. It is a modified G.729 heart-shaped hierarchical encoder producing a bandwidth ranging from narrow band (50-4000 Hz) to wide band (50-7000 Hz) at a rate of 8 to 32 kbit / s for conversational services. This codec is compatible with existing VoIP devices that use the G.729 codec.

The G.729.1 coder is shown diagrammatically in FIG. 1. The broadband input signal s, sampled at 16 kHz, is first decomposed into two subbands by QMF (for "Quadrature Mirror Filter") filtering. The low band (0-4000 Hz) is obtained by LP low-pass filtering (block 100) and decimation (block 101), and the band high (4000-8000 Hz) by high-pass filtering HP (block 102) and decimation (block 103). The LP and HP filters are of length 64.

The low band is pretreated with a high-pass filter eliminating the components below 50 Hz (block 104), to obtain the signal S ₁₈ , before CELP coding in narrow band (block 105) at 8 and 12 kbit / s. This high-pass filtering takes into account the fact that the wanted band is defined as covering the interval 50-7000 Hz. The narrow-band CELP coding is a cascaded CELP coding comprising as a first stage a modified G.729 coding without a filter. preprocessing and as a second stage an additional fixed CELP dictionary.

 The high band is first pretreated (block 106) to compensate for the folding due to the high-pass filter (block 102) combined with the decimation (block 103). The high band is then filtered by a low pass filter (block 107) eliminating the components between 3000 and 4000 Hz from the high band (that is, the components between 7000 and

8000 Hz in the original signal) to obtain the _HB signal. A parametric band extension (block 108) is then performed.

An important feature of the G.729.1 encoder according to Figure 1 is as follows. The error signal d _LB of the low band is calculated (block 109) from the output of the CELP coder (block 105) and a predictive coding by transform (of type

TDAC for "Time Domain Aliasing Cancellation" in the G.729.1 standard) is carried out at block 110. With reference to FIG. 1, it can be seen in particular that the encoding

TDAC is applied to both the error signal on the low band and the signal filtered on the high band.

Additional parameters may be transmitted by the block 1 11 to a homologous decoder, this block 111 performing a so-called "FEC" treatment for "Frame Erasure Concealment", in order to reconstruct any erased frames. - AT -

The different bit streams generated by the coding blocks 105, 108, 110 and 111 are finally multiplexed and structured into a hierarchical bit stream in the multiplexing block 112. The coding is performed by 20 ms sample blocks (or frames). 320 samples per frame.

 The coded G .129.1 therefore has a three-step coding architecture comprising: CELP coding in cascade,

 the parametric band extension by the module 108, of the TDBWE (for "Time Domain Bandwidth Extension") type, and

 a TDAC transform predictive coding, applied after an MDCT (Modified Discrete Cosine Transform) transformation.

* Reminders on the G.729.1 decoder The G.729.1 decoder is illustrated in Figure 2. The bits describing each frame of 20 ms are demultiplexed in the block 200.

The bit stream of the 8 and 12 kbit / s layers is used by the CELP decoder (block 201) to generate the narrow-band synthesis (0-4000 Hz). The portion of the bit stream associated with the 14 kbit / s layer is decoded by the tape extension module (block 202). The portion of the bit stream associated with data rates greater than 14 kbit / s is decoded by the TDAC module (block 203). Pre-echo and post-echo processing is performed by blocks 204 and 207 as well as enrichment (block 205) and aftertreatment of the low band (block 206). The extended band output signal s _wb , sampled at 16 kHz, is obtained via the QMF synthesis filter bank (blocks 209, 210, 211, 212 and 213) integrating the inverse folding (block 208).

 The description of the transform coding layer is detailed below.

* Feedback on the TDAC Transform Encoder in the G.729.1 Encoder

The TDAC type transform coding in the G.729.1 encoder is illustrated in FIG.

The filter W ₁₈ (Z) (block 300) is a perceptual weighting filter, with gain compensation, applied to the low band error signal d _LB. MDCT transforms are then calculated (blocks 301 and 302) to obtain:

the MDCT spectrum D ^ of the difference signal, perceptually filtered, and the spectrum MDCT S _HB of the original signal of the high band. These MDCT transforms (blocks 301 and 302) apply to 20 ms of sampled signal at 8 kHz (160 coefficients). The spectrum Y (k) from the block 303 of fusion thus comprises 2 x 160, or 320 coefficients. It is defined as follows:

[F (O) F (I) - - 7 (319)] = [D _L ^w _B (O) D 1 (I) -D 1 (159) S _HB (O) S _HB (1) -5 _Wfl ( 159) J

 This spectrum is divided into eighteen sub-bands, a sub-band j being assigned a number of coefficients noted nb _coef (j). The subband splitting is specified in Table 1 below.

Thus, a subband j comprises the coefficients Y (k) with sb _ bound (j) ≤ k <sb _ bυund (j + 1). It should be noted that the coefficients 280-319 corresponding to the frequency band 7000 Hz - 8000 Hz are not coded; they are set to zero at the decoder because the bandwidth of the codec is 50-7000 Hz.

Table 1: Limits and size of subbands in TDAC coding The spectral envelope {\ og_rms (j)} _χη is calculated in block 304 according to the formula: Y (k) ² + ε _rm j = 0, ..., n

where £ ,, = 2 -24

The spectral envelope is coded at variable rate in block 305. This block 305 produces quantized values, integer, denoted by _index (j) (with 7 = 0, ..., 17), obtained by simple scalar quantization:

 set _ index (j) = round (2 • log_rms (j)) where the notation "round" designates the rounding to the nearest integer, and with the constraint:

 - I I <put _ index (j) ≤ +20

 This quantized value set _index (j) is transmitted to the bit allocation block 306.

 The coding of the spectral envelope, itself, is carried out again by the block 305, separately for the low band (set _index (j), with 7 = 0, ..., 9) and for the high band (set _index (j), with 7 = 10, ..., 17). In each band, two types of coding can be chosen according to a given criterion, and, more precisely, the values put _index (j):

 can be coded by "differential Huffman" coding,

 or can be encoded by natural binary coding.

 A bit (0 or 1) is transmitted to the decoder to indicate the encoding mode that has been chosen.

 The number of bits allocated to each subband for its quantization is determined at block 306 from the quantized spectral envelope from block 305.

The bit allocation performed minimizes the squared error while respecting the constraint of a whole number of bits allocated per subband and a number of bits. maximum not to be exceeded. The spectral content of the subbands is then encoded by spherical vector quantization (block 307).

The different bit streams generated by the blocks 305 and 307 are then multiplexed and structured into a hierarchical bit stream at the multiplexing block 308.

* Reminder on the decoder by transform in the G.729.1 decoder

The TDAC-type transform decoding step in the G.729.1 decoder is illustrated in FIG. 4.

 In a symmetrical way to the encoder (FIG. 3), the decoded spectral envelope

(block 401) allows to find the allocation of bits (block 402). Envelope decoding (block 401) reconstructs the quantized values of the spectral envelope

(set _index (j), for j = 0, ..., 17), from the bit stream generated by block 305 (multiplexed) and deduces from it the decoded envelope:

rms _ q (j) = 2 ' ^{A ms} -' ^ofUj)

 The spectral content of each of the subbands is found by inverse spherical vector quantization (block 403). The non-transmitted sub-bands, due to a lack of "budget" of bits, are extrapolated (block 404) from the MDCT transform of the signal at the output of the band extension block (block 202 of FIG. 2).

 After upgrading this spectrum (block 405) as a function of the spectral envelope and after-treatment (block 406), the MDCT spectrum is separated into two (block 407):

with 160 first coefficients corresponding to the spectrum D i of the decoded difference signal in low band, filtered perceptually, and 160 subsequent coefficients corresponding to the spectrum S _HB of the original decoded high band signal.

These two spectra are transformed into time signals by inverse MDCT transform, denoted IMDCT (blocks 408 and 410), and the inverse perceptual weighting (filter denoted W ₁₈ (Z) ^'1 ) is applied to the signal d _L ^w _B (block 409 ) resulting from the inverse transform.

 The sub-bit allocation (block 306 of FIG. 3 or block 402 of FIG. 4) is described more particularly below.

 Blocks 306 and 402 perform an identical operation from the values set _index (j), 7 = 0, ..., 17. It is therefore merely necessary to describe only the operation of block 306.

 The purpose of the binary allocation is to distribute between each of the sub-bands a certain bit budget (variable) noted nbits _ VQ, with:

nbits _VQ = 351 - nbits _ put, where nbits _ mis is the number of bits used by the coding of the spectral envelope.

 The result of the allocation is the integer number of bits, denoted nbit (j) (with = 0, ..., 17), allocated to each of the sub-bands with as global constraint:

 17

 Σnbiti j) ≤ nbits _VQ

y =

In the G.729.1 standard, the nbit (j) values (/=0,...,17) are further constrained by the fact that nbit (j) must be selected from a reduced set of values specified in Table 2 below.

Table 2: Possible values of number of bits allocated in TDAC subbands.

 The allocation in G.729.1 is based on a "perceptual importance" per subband related to the energy of the subband, denoted ip (j) (/=0..17), defined as follows:

ψ (j) - -Iog ₂ (rms _q (jf xnb _coef (j)) + offset

 where offset = -2.

Since the rms values _q (j) = 2 ^{'to R' m} - ^{mJex (J)} _{^ e} ^ _this formula simplifies as: _ rms index (j) for j = 0, ..., 16

 ip (j) =

 (rms _ index (j) - l) for j = 17

From the perceptual importance of each sub-band, the allocation nbit (j) is calculated as follows:

nbit (j) = arg min nb _coef (j) x (ip (j) - λ) - r where λ is a dichotomy optimized parameter to satisfy the global constraint 17

 Σnbit (j) ≤nbits _VQ

 J = O

approaching the nbits_VQ threshold.

The incidence of perceptual weighting (filtering of block 300) on the bit allocation (block 306) of the TDAC transform coder is described in more detail below.

In the G.729.1 standard, the TDAC coding uses the perceptual weighting W _{/ B} (z) filter in the low band (block 300), as indicated above. In essence, perceptual weighting filtering allows you to format the coding noise. The principle of this filtering is to exploit the fact that it is possible to inject more noise in the frequency zones where the original signal has a high energy.

 The most common perceptual weighting filters used in narrow-band CELP coding are of the form  (z / γ1) /  (z / γ2) where 0 <γ2 <γl <1 and  (z) represents a prediction spectrum linear (LPC). The synthesis analysis in CELP coding thus amounts to minimizing the quadratic error in a signal domain perceptually weighted by this type of filter.

However, to ensure the spectral continuity when the spectra D " _B and S _ιm are contiguous (block 303 of FIG. 3), the filter W _LB (z) is defined in the form:

with γ, The fac factor makes it possible to ensure at the junction of the low and high bands (4 kHz) a gain of the filter at 1 to 4 kHz. It is important to note that in the G.729.1 TDAC coding, the coding is based on an energetic criterion only. * Disadvantages of the prior art

 The energy criterion of TDAC coding of G.729.1, used in the high band (4000-7000 Hz), is not optimal from a perceptual point of view, in particular to code musical signals.

 The perceptual weighting filter is particularly suited to speech signals. It is widely used in speech coding standards based on the CELP coding format. However, for the musical signals, it appears that this perceptual weighting based on a shaping of the quantization noise according to the formants of the input signal is insufficient. Most audio coders rely on transform coding using frequency masking or simultaneous masking models; they are more generic (in the sense that they do not use a speech production model like the CELP) and are therefore more suitable for encoding musical signals.

 Reference can be made to Mr. Bosi and R. Goldberg's "Introduction to digital audio coding and standards," published by Kluver Academy Publishers, in 2003, for more details on masking patterns and their application in transform coders.

 There is therefore a need to improve the signal coding quality for better perceptual rendering, while maintaining interoperability with G.729.1 coding.

The present invention improves the situation.

It proposes for this purpose, a method of hierarchical coding of a digital audio input signal in several frequency subbands comprising heart encoding of the input signal at a first rate and at least one higher rate enhancement coding of a residual signal, the core coding using a binary allocation according to an energy criterion. The method is such that it comprises the following steps for improvement coding:

 calculating a frequency masking threshold for at least part of the frequency bands processed by the improvement coding;

 determining perceptual importance by sub-frequency band according to the calculated masking threshold and as a function of the number of bits allocated for the core coding;

 binary bit allocation in the frequency subbands processed by the enhancement coding, according to the determined perceptual importance; and

 -coding the residual signal according to the bit allocation.

 Thus, the coding according to the invention takes advantage of an enhancement coding layer to improve the coding quality from a perceptual point of view. The enhancement layer will thus benefit from frequency masking which does not exist in the core coding stage, in order to best allocate the bits in the frequency bands of the improvement coding.

 This does not change the core coding, which is thus compatible with existing standard coding, thus ensuring interoperability with existing equipment that uses existing standard coding.

 The various particular embodiments mentioned below may be added independently or in combination with each other, to the steps of the coding method defined above.

In a particular embodiment, the step of determining a perceptual importance comprises: a first step of defining a first perceptual importance for at least one frequency sub-band of the improvement coding, as a function of the frequency masking threshold in the sub-band, of quantized values of the coding of the spectral envelope for the frequency subband and a determined normalization factor;

 a second step of subtracting from the first perceptual importance a ratio between the number of bits allocated for the core coding and the number of coefficients in said sub-band.

 Thus, the first perceptual importance that will be used for the improvement layer, does not take into account the core coding but only the signal to mask ratio to define a perceptual importance. This perceptual importance is determined on the input signal of the transform coder.

 Taking into account the core coding is simply done by subtracting the average number of bits per sample already allocated. The use of the perceptual importance based on the signal-to-mask ratio would make it possible to obtain an optimal allocation in the perceptual sense. However, this allocation would be useful if the input signal of the transform coding layer was coded directly. In the context of the invention, a first transform coding layer based on an energy allocation has allocated a certain number of bits per subband.

If one wants to improve the quality by coding the residual signal of this core encoder layer without wasting rate, it is necessary to adapt the perceptual importance based on the signal to mask ratio of the input signal to the residual signal. For this purpose, a value representative of the number of bits allocated in the core coder is subtracted from the first perceptual importance. It should be noted that the perceptual importance based on the signal to mask ratio of a residual signal can not be calculated. Indeed, in this case the masking curve that would be calculated would not really have a perceptive meaning, since it would not be based on the perceived signal.

 In a variant embodiment, the perceptual importance is furthermore determined as a function of bits allocated for coding of improvement of the preceding core coding, having a binary allocation according to an energy criterion.

 In the G.729.1 decoder the non-transmitted subbands, due to a lack of sufficient bit budget, are extrapolated (block 404) from the MDCT transform of the output signal of the band extension block (block 202 of FIG. ). Even at the higher bit rate of G.729.1 (32 kbit / s) some frequency bands remain extrapolated. Before applying the enhancement coding according to the present invention, it is first possible to use a first coding enhancement coding heart to fill the heart encoding bit rate for these non-transmitted subbands. This first enhancement coding uses the original signal and operates according to energy criteria for bit allocation. According to one embodiment of the invention, this first enhancement coding modifies the number of bits nbi 1 (j) allocated to the subbands and the decoded subband Y q (k) (defined later in FIG. 5).

 The improvement coding according to the invention therefore also takes into account the bits allocated during this first improvement coding, in addition to the bits allocated in the core coding.

 Advantageously, the masking threshold is determined for a sub-band, by a convoi ution between:

 an expression of a calculated spectral envelope, and

a spreading function involving a central frequency of said subband. In an alternative embodiment, the method comprises a step of obtaining an information according to which the signal to be encoded is tonal or non-tonal and the steps of calculating the masking threshold and of determining a perceptual importance according to this masking threshold, are conducted only if the signal is non-tonal.

 Thus, the coding is adapted to the signal whether it is tonal or not and allows optimal allocation of the bits.

 In a particularly suitable application of the invention, the improvement coding is a TDAC type enhancement coding in an extended coder whose core coding is of the G.729.1 standard encoder type.

 Thus, the quality of the G.729.1 codec in the enlarged band (50-7000 Hz) is improved. Such an improvement is important to extend the G.729.1 encoder band of the enlarged band (50-7000Hz) to the super-wide band (50-14000Hz).

 The present invention also relates to a method of hierarchical decoding of a digital audio frequency signal into a plurality of frequency subbands comprising a core decoding of a signal received at a first rate and at least a higher rate of improvement decoding, d a residual signal, the core decoding using a binary allocation according to an energy criterion. The method is such that it comprises the following steps for the improvement decoding:

 calculating a frequency masking threshold for at least a portion of the frequency subbands processed by the enhancement decoding;

 determination of a perceptual importance by frequency sub-band according to the calculated masking threshold and as a function of the number of bits allocated for the core decoding;

 bit allocation in the frequency sub-bands processed by the enhancement decoding, according to the determined perceptual importance; and

decoding the residual signal according to the bit allocation. In the same way and with the same advantages as for coding, the step of determining a perceptual importance comprises:

 a first step of defining a first perceptual importance for at least one frequency sub-band of the improvement decoding, as a function of the frequency masking threshold in the sub-band, of quantized values of the spectral envelope decoding for the frequency sub-band and a determined normalization factor;

 a second step of subtracting from the first perceptual importance a ratio between the number of bits allocated for the core decoding and the number of coefficients in said subband.

 The invention relates to a hierarchical coder of a digital audio frequency input signal in several frequency subbands comprising a core encoder of the input signal according to a first bit rate and at least one higher bit rate improvement coder, d a residual signal, the core coder using a binary allocation according to an energy criterion. The enhancement coder includes:

 a module for calculating a frequency masking threshold for at least part of the frequency bands processed by the improvement coder;

 a module for determining perceptual importance by sub-frequency band according to the calculated masking threshold and as a function of the number of bits allocated for the core coder;

 a module for bit binary allocation in the frequency sub-bands processed by the improvement coder, according to the determined perceptual importance; and

 a module for coding the residual signal according to the bit allocation.

It also relates to a hierarchical decoder of a digital audio signal in several frequency subbands including a heart decoder of a signal received at a first rate and at least one higher rate improvement decoder, of a residual signal, the heart decoder using a binary allocation according to an energy criterion. The enhancement decoder comprises:

 a module for calculating a frequency masking threshold for at least a portion of the frequency sub-bands processed by the improvement decoder;

 a module for determining a perceptual importance per frequency subband as a function of the calculated masking threshold and as a function of the number of bits allocated for the heart decoder;

 a bit allocation module in the frequency sub-bands processed by the enhancement decoder, according to the determined perceptual importance; and

 a module for decoding the residual signal according to the allocation of bits.

 Finally, the invention relates to a computer program comprising code instructions for implementing the steps of a coding method according to the invention, when they are executed by a processor and to a computer program comprising instructions. code for implementing the steps of a decoding method according to the invention, when executed by a processor.

 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:

 FIG. 1 illustrates the structure of a G.729.1 type encoder described above;

FIG. 2 illustrates the structure of a G.729.1 type decoder described above; FIG. 3 illustrates the structure of a TDAC encoder included in the G.729.1 type encoder and described previously:

 FIG. 4 illustrates the structure of a TDAC decoder included in a G.729.1 decoder and as described above;

 FIG. 5 illustrates the structure of a TDAC coder comprising enhancement coding according to one embodiment of the invention;

 FIG. 6 illustrates the structure of a TDAC decoder comprising improvement decoding according to one embodiment of the invention;

 FIG. 7 illustrates an advantageous spreading function for masking within the meaning of the invention;

 FIG. 8 illustrates a normalization of the masking curve, in one embodiment of the invention;

 FIG. 9 illustrates the structure of a frequency band extended G.729.1 encoder in which a TDAC encoder according to one embodiment of the invention is included;

 FIG. 10 illustrates the structure of a G.729.1 extended frequency band decoder in which a TDAC decoder according to one embodiment of the invention is included;

 FIG. 11a illustrates an exemplary hardware embodiment of a terminal including an encoder according to an embodiment of the invention; and

 FIG. 11b illustrates an example of a hardware embodiment of a terminal including a decoder according to one embodiment of the invention.

One of the objects of the invention is the improvement of the quality of G.729.1 in wideband (50-7000 Hz), in particular for musical signals. It will be recalled here that the G.729.1 coding has a useful band of 50 to 7000 Hz. Moreover, the quality of G.729.1 for some signals such as music signals is not transparent at its highest bit rate (32 kbit / s) - this limitation is due to the hierarchical structure CELP + TDBWE + TDAC and the bit rate limited to 32 kbit / s .

 This invention is motivated by the ongoing standardization in ITU-T of a scalable extension of G.729.1, in particular to extend the coded band by

G.729.1 to the super-enlarged band (50-14000 Hz). Experience shows that the band extension (eg 7000-14000 Hz) of a limited band signal (eg 50-7000 Hz) requires having a limited band signal that is already of good quality; indeed the band extension highlights the existing defects in this signal. Thus, there is a need to improve the quality of G.729.1 in broadband (50-7000 Hz).

 The G.729.1 quality improvement can be achieved with one or more additional rate enhancement layers (in addition to 32 kbit / s). In practice these additional flow enhancement layers can serve both the band extension (7000-14000 Hz) and the quality improvement in the enlarged band (50-7000 Hz). Thus, part of the additional bit rate of the enhancement layers can be devoted to improving the broadband signal decoded by a G.729.1 decoder.

 It should be noted that two cores can be distinguished in the hierarchical coding considered in this document: G.729.1 has a narrow-band CELP core coder, while the super-expanded band extension (50-14000Hz) of G.729.1 has for heart G.729.1.

 In the following, the terms core and core rate coding mean a G.729.1 type coding and the associated bit rate of 32 kbit / s.

In one embodiment of the invention, it is more particularly a TDAC encoder and decoder as described above, in which an enhancement layer is integrated. Figure 5 describes such an improved TDAC coder.

 A scalable extension of G.729.1 is considered in several enhancement layers. Here the core coding is a G.729.1 coding, which uses TDAC coding in the band [50-7000 Hz] from the bit rate of 14 kbit / s to 32 kbit / s. It is assumed that between 32 and 48 kbit / s two enhancement layers of 8 kbit / s are produced in order to extend the band from 7000 to 14000 Hz and to replace the non-transmitted subbands of the TDAC coding of G.729.1 . These enhancement layers of 8 kbit / s ranging from 32 to 48 kbit / s are not described here.

 The present invention provides two additional 8 kbit / s enhancement layers of the TDAC coding in the 50 to 7000 Hz band that increase the bit rate from 48 kbit / s to 56 and 64 kbit / s.

 The encoder applying the present invention has enhancement layers that add G.729.1 core bit rate (32 kbit). These enhancement layers serve both to improve the quality in the enlarged band (50-7000 Hz) and to extend the upper band from 7000 to 14000 Hz. In the following we ignore the extension of 7000 to 14000 Hz, because this feature does not influence the implementation of the present invention. For reasons of simplicity, the modules corresponding to the 7000 to 14000 Hz band extension are not illustrated in FIGS. 5 and 6.

 Here we find the same blocks (blocks 500 to 507) as those used in the base layer of G.729.1 (blocks 300 to 307) as described with reference to FIG.

 The TDAC encoder according to one embodiment of the invention here comprises an enhancement layer (blocks 509 to 513) which improves the core layer (blocks 504 to 507).

Note that the block 507 here corresponds to the spherical vector quantization (SVQ) of G.729.1, which may comprise a modification as mentioned above. Thus, in this block 507, a first coding for improving the G.729.1 core coding is used to fill the lack of bit rate for the non-transmitted subbands (where nbit (j) = 0). This modification uses the original signal Y (k) and operates according to energy criteria for bit allocation. The number of bits nbit (j) allocated to the subbands and the decoded subband Yq (k) are then modified.

 Block 506 performs a binary allocation based on energy criteria as described with reference to FIG.

 The core layer is therefore coded and sent to the multiplexing module 508. The core signal is also decoded locally in the coder by the block.

510 which performs spherical dequantization and scaling; this heart signal is subtracted from the original signal at 509, in the transformed domain, to obtain a residual signal err (k). This residual signal is then coded from a bit rate of 48 kbit / s in block 513.

Block 511 computes a masking curve from the coded spectral envelope rms _ q (j) obtained by block 505, where y = 0, ..., 17 is the number of the subband.

The masking threshold M (j) of the subband j is defined by the convolution of the energy envelope at ² (j) = rms_q (j) ² xnb _coef (j), by a function of spreading B (v).

 In a first embodiment, this masking is performed only on the high band of the signal, with:

M ( _J ) =] T σ ² (/ :) x 5 (v ₇ -v,)

/ (= 10 where v _k is the central frequency of the sub-band k in Bark,

 the sign "x" denoting "multiplied by", with the spreading function described below.

 In more generic terms, the masking threshold M (J) for a sub-band j is therefore defined by a convolution between:

 an expression of the spectral envelope, and

 a spreading function involving a central frequency of the sub-band

An advantageous spreading function is that shown in FIG. 7. It is a triangular function whose first slope is + 27dB / Bark and -10dB / Bark for the second. This representation of the spreading function allows the iterative calculation of the following masking curve:

 M- (IO) 7 = 10

M {j) = M ⁺ (j) + M ^~ (j) + σ ² (j) j = 1 .., 16

M ⁺ (II) J = IV

or

MHj) = ² O- (JI) - A ₂ (J) ₊ M ⁺ (JI) - A ₂ (J) y = 11, .., 17 M ^U) = ² (j + l) - A _ι (j) + M ^~ (J ₊ I) - A ₁ (J) 7 = 10, .., 16 and

-10 _/

Δ ₂ (./) = 10 ^"

A ₁ (J) = IO Έ 10 ("> - ^V > * ^< )

The values of A ₁ (J) and Δ ₉ (j) can be pre-calculated and stored. Since the low band is already filtered perceptually by the module 500, the application of the masking threshold is in this embodiment, limited to the high band. In order to ensure the spectral continuity between the low and high band spectrum weighted by the masking threshold and to avoid biasing the bit allocation, the masking threshold is normalized for example by its value on the last sub-band. band of the low band.

A first perceptual importance calculation step is then performed taking into account the signal-to-mask ratio given by:

 The perceptual importance is therefore defined as follows in block 511: 0..9 10..17

where offset = -2 and normfac is a normalization factor calculated according to the relation: normfac = log ₂ JT σ ² (j) x B (V ₉ - V _j )

 y = 9

 We note that the perceptual importance ip (j), j = 0, ..., 9, is identical to that defined in the G.729.1 standard. On the other hand, the definition of the term ip (j), j = lθ, ..., l7, is changed.

 The perceptual importance defined above is now written:

- put _ index (j) for 7 = 0, ..., 9

 ip (j) =

- [rms _ index (j) - log_ mask (j)] for j = 10, ..., 17 where log_ mask (j) = log ₂ (M (J)) - normfac.

 An illustration of the standardization of the masking threshold is given in FIG. 8, showing the connection of the high band on which the masking (4-7 kHz) is applied to the low band (0-4 kHz).

In a variant of this embodiment in which the masking threshold is normalized with respect to its value on the last subband of the low band, the standardization of the masking threshold can be rather carried out from the value of the band. masking threshold in the first subband of the high band, as follows: normfac = log ₂ Σσ ² α) x5 (v _l0 -v _y )

 y = 10

In another variant, the masking threshold can be calculated over the entire frequency band, with:

The masking threshold is then applied only to the high band after normalization of the masking threshold by its value on the last subband of the low band: normfac = log ₂ Σσ ² (j) x B (V ₉ -V ₁ )

 J = O

or else by its value on the first subband of the high band: normfac = log ₂ Σσ ² (j) x B (V ₁₀ - v _; )

 j = 0

Of course, these relations giving the normalization factor normfac or the masking threshold M (J) can be generalized to any number of subbands (different, in total, from eighteen) in high band (with a different number of eight), as in low band (with a different number of ten).

 From this frequency masking calculation, a first perceptual importance ip (j) is sent to the binary allocation block 512 for improvement coding.

 This block 512 also receives bit allocation information nbit (j) from the core layer of the TDAC coding, G.729.1.

 Block 512 thus defines a new perceptual importance that takes into account these two pieces of information.

So, a second perceptual importance is defined as follows: ip '(Jl = ⁽ PU) - " ^{blt (j} 2,., Pourj = l, ..., 18

nb_coeff ⁽ j ⁾

where nbit (j) represents the number of bits allocated by the base layer to the frequency band j, and nb_coeff (j) represents the number of coefficients of the band j according to Table 1 described above.

 In other words, the new perceptual importance is calculated by subtracting from the first perceptual importance, a ratio between the number of bits allocated for core coding and the number of possible coefficients in the subband.

 With this new perceptual importance, the block 512 performs a bit allocation on the residual signal to code the enhancement layer.

 This bit allocation is calculated as follows:

nbit_err (j) = arg _{reR t} At _ιψl ) - r where optimization must satisfy the constraint

 17

 ^ nbit _ err (j) ≤ nbits _ VQ _ err

j = 0 nbits _VQ_err corresponding to the additional number of bits in the enhancement layer (320 bits for the 2 8 kbit / s layers).

 It therefore takes into account the new calculated perceptual importance.

The residual signal err (k) is then coded by the module 513 by spherical vector quantization, using the number of allocated bits nbit_err (j) as previously calculated.

 This coded residual signal is then multiplexed with the signal resulting from the core coding and the coded envelope by the multiplexing module 508.

This enhancement coding, not only extends the allocated bit rate but improves from a perceptual point of view, the coding of the signal.

 It is recalled that the TDAC coding enhancement layer as described can be applied after modifying the TDAC coding of G.729.1. In the 32 to 48 kbit / s enhancement layers, a first enhancement (not described here) of the TDAC coding of G.729.1 is performed. This enhancement allocates bits to subbands between 4 and 7 kHz where no bit rate was allocated by the G.729.1 TDAC core encoding even at its higher 32 kbit / s bit rate. This first improvement of the TDAC coding of G.729.1 therefore uses the original signal between 4 and 7 kHz and does not implement the steps of calculating a masking threshold nor determining the perceptual importance of the coding method of the 'invention. Block 507 is considered to correspond to this modified TDAC coding integrating this improvement.

 Thus, in the enhancement layer of the coding method of the invention, at rates ranging from 48 kbit / s to 64 kbit / s, the determination of the perceptual importance

(blocks 511, 512) not only takes into account the bits allocated for core or base coding but also the bits allocated for the previous enhancement coding, in this case, the 40 kbit / s rate enhancement coding . FIG. 5 not only illustrates the TDAC coder with its improvement coding stage but also serves to illustrate the steps of the coding method according to one embodiment of the invention as described above and in particular the steps of:

calculating a frequency masking threshold for at least part of the frequency bands processed by the improvement coding;

 determination of a perceptual importance by frequency sub-band according to the calculated masking threshold and as a function of the number of bits allocated for the core coding;

binary bit allocation in the frequency subbands processed by the enhancement coding, according to the determined perceptual importance; and

 coding of the residual signal according to the bit allocation.

 FIG. 6 illustrates the TDAC decoder with an enhancement decoding stage as well as the steps of a decoding method according to one embodiment of the invention.

 The decoder comprises the modules (601, 602, 603, 606, 607, 608, 609 and 610) identical to those described for the TDAC decoding of the G.729.1 coder with reference to FIG. 4 (401, 402, 403, 406, 407, 408, 409 and 410). Note that the block 606 for processing in the MDCT domain (aimed at shaping the coding noise) is here optional because the invention improves the quality of the decoded MDCT spectrum from block 603.

 The module 605 of the decoder corresponds to the encoder module 511 and operates in the same way from the quantized values of the spectral envelope.

From the first perceptual importance ip (j) calculated by this module 605, the allocation module 604 determines a second perceptual importance in taking into account the allocation of bits received from the core coding, in the same way as in the coding module 512.

 This bit allocation for the enhancement coding allows the module 611 to decode the signal received from the demultiplexing module 600 by spherical vector dequantization.

 The decoded signal from the module 611 is an error signal err (k) which is then combined at 612 with the decoded heart signal at 603.

This signal is then processed as for the G.729.1 coding described with reference to FIG. 4, to give a difference signal d _L g in low band and a signal S _HB in high band.

 It is also indicated that the calculation of a frequency masking performed by the module 511 or 605 and as described above, may or may not be carried out according to the signal to be encoded (in particular if it is tonal or not).

 It has indeed been observed that the calculation of the masking threshold is particularly advantageous when the signal to be encoded is not tonal.

 If the signal is tonal, the application of the spreading function B (v) results in a masking threshold very close to a tone a little more spread out in frequencies. The criterion of minimization of the masked coding noise ratio then gives a bit allocation that is not necessarily optimal.

 It is therefore possible to improve this allocation, use a bit allocation according to energy criteria for a tonal signal.

 Thus, in an alternative embodiment, the calculation of the masking threshold and the determination of the perceptual importance as a function of this masking threshold according to the invention is applied only if the signal to be encoded is not tonal.

In generic terms, we obtain information (from block 505) that the signal to be encoded is tonal or non-tonal, and the perceptual weighting of the high band, with the determination of the masking threshold and normalization, are conducted only if the signal is non-tonal.

 With G.729.1 type core coding, the bit relating to the mode of the spectral envelope coding (block 505 or 601) indicates a "differential Huffman" mode or a "natural direct binary" mode. This mode bit can be interpreted as a tone detection, since, in general, a tonal signal leads to envelope coding by the "natural direct binary" mode, while most non-tonal signals, having a spectral dynamic more limited, lead to envelope coding by the "Differential Huffman" mode.

 Thus, it can be taken advantage of the "signal tone detection" to implement frequency masking or not. More particularly, this masking threshold calculation is applied in the case where the spectral envelope has been coded in "differential Huffman" mode and the first perceptual importance is defined in the sense of the invention, as follows:

On the other hand, if the envelope has been coded in "natural direct binary" mode, the first perceptual importance remains as defined in the G.729.1 standard:

0, ..., l6

 17

A possible application of the invention to an extension of the G.729.1 encoder, in particular in super-wideband, is now described.

 With reference to FIG. 9, such an encoder is illustrated. The super-wideband extension of the G.729.1 encoder as shown consists of an extension of the frequencies coded by the module 915, the frequency band used from

[50Hz-7KHz] to [50Hz-14kHz] and an improvement of the base layer of the

G.729.1 by the TDAC coding module (block 910) and as described with reference to FIG. 5.

 Thus, the encoder as represented in FIG. 9 comprises the same modules as the core G.729.1 coding shown in FIG. 1 and an additional band extension module 915 which provides an extension signal to the multiplexing module 912.

This frequency band extension is calculated on the original full-band signal S _SWB while the input signal of the core encoder is obtained by decimation (block 913) and low-pass filtering (block 914). At the output of these blocks, the broadband input signal S _WB is obtained.

 The TDAC coding module 910 is different from that illustrated in FIG. 1. This module is for example that described with reference to FIG. 5 and provides the multiplexing module with both the coded core signal and the coded improvement signal. the invention.

 In the same way, a G.729.1 decoder extended in super-wideband is described with reference to FIG. 10. It comprises the same modules as the decoder

G.729.1 described with reference to FIG.

However, it comprises an additional band extension module 1014 which receives from the demultiplexing module 1000, the band extension signal. It also comprises the synthesis filter bank (blocks 1015, 1016) making it possible to obtain the super-wideband output signal S _swh .

 The TDAC decoding module 1003 is also different from the TDAC decoding module illustrated with reference to FIG. 2. This module is for example that described and illustrated with reference to FIG. 6. It therefore receives from the demultiplexing module, both the heart signal and the signal of improvement.

 In the preferred embodiment presented above, the invention is used to improve the quality of TDAC coding in the G.729.1 codec. Naturally, the invention applies to other types of transform coding with a binary allocation and to the scalable extension of other core codes than G.729.1.

 An exemplary hardware embodiment of the encoder and the decoder as described with reference to FIGS. 5 and 6 is now described with reference to FIGS. 11a and 11b.

 Thus, FIG. 11a illustrates an encoder or terminal comprising an encoder as described in FIG. 5. It comprises a processor PROC cooperating with a memory block BM comprising a storage and / or working memory MEM.

This terminal comprises an input module able to receive a low band signal d _LB and a high band signal S _HB or any type of digital signal to be coded. These signals may come from another coding stage or a communication network, a digital content storage memory.

 The memory block BM may advantageously comprise a computer program comprising code instructions for implementing the steps of the coding method in the sense of the invention, when these instructions are executed by the processor PROC, and in particular the steps of:

calculating a frequency masking threshold for at least a portion of the frequency sub-bands processed by the improvement coding; determination of a perceptual importance by frequency sub-band according to the calculated masking threshold and as a function of the number of bits allocated for the core coding;

 bit allocation in the frequency sub-bands processed by the improvement coding, according to the determined perceptual importance; and

 coding of the residual signal according to the bit allocation.

 Typically, the description of FIG. 5 repeats the steps of an algorithm of such a computer program. The computer program can also be stored on a memory medium readable by a reader of the terminal or encoder or downloadable in the memory space thereof.

 The terminal comprises an output module capable of transmitting a multiplexed stream derived from the coding of the input signals.

 In the same way, FIG. 11b illustrates an example of a decoder or terminal including a decoder as described with reference to FIG.

 This terminal comprises a PROC processor cooperating with a memory block

BM having a memory storage and / or working MEM.

 The terminal comprises an input module adapted to receive a multiplexed stream coming for example from a communication network, a storage module.

 The memory block may advantageously comprise a computer program comprising code instructions for implementing the steps of the decoding method in the sense of the invention, when these instructions are executed by the processor PROC, and in particular the steps of:

calculating a frequency masking threshold for at least a portion of the frequency subbands processed by the enhancement decoding; determination of a perceptual importance by frequency sub-band according to the calculated masking threshold and as a function of the number of bits allocated for the core decoding;

 bit allocation in the frequency sub-bands processed by the enhancement decoding, according to the determined perceptual importance; and

decoding the residual signal according to the bit allocation.

 Typically, the description of FIG. 6 shows the steps of an algorithm of such a computer program. The computer program can also be stored on a memory medium readable by a reader of the terminal or downloadable in the memory space thereof.

The terminal comprises an output module able to transmit decoded signals {dtB, S _HB ) for another coding stage or for a content reproduction.

 Of course, such a terminal may comprise both the encoder and the decoder according to the invention.

Claims

1. A method of hierarchical coding of an input digital audio frequency signal in several frequency sub-bands comprising a core encoding of the input signal according to a first bit rate and at least one higher bit rate improvement coding of a residual signal. , the core coding using a binary allocation (506) according to an energy criterion, characterized in that it comprises the following steps for the improvement coding:

 calculating a frequency masking threshold (511) for at least a portion of the frequency bands processed by the improvement coding;

 determination (511, 512) of a perceptual importance per frequency sub-band as a function of the calculated masking threshold and as a function of the number of bits allocated for the core coding;

 bit allocation (512) of bits in the frequency sub-bands processed by the enhancement coding, according to the determined perceptual importance; and

-coding the residual signal (513) according to the bit allocation.

2. Method according to claim 1, characterized in that the step of determining a perceptual importance comprises:

a first step (511) for defining a first perceptual importance for at least one frequency sub-band of the improvement coding, as a function of the frequency masking threshold in the sub-band, of quantified values of the coding of the spectral envelope for the frequency subband and a determined normalization factor; a second step (512) of subtracting from the first perceptual importance a ratio between the number of bits allocated for the core coding and the number of coefficients in said sub-band.

3. Method according to claim 1, characterized in that the perceptual importance is further determined as a function of bits allocated for coding for improving the previous core coding, having a binary allocation according to an energy criterion.

4. Method according to claim 1, characterized in that the masking threshold is determined for a subband, by a convolution between:

 an expression of a calculated spectral envelope, and

 a spreading function involving a central frequency of said subband.

5. Method according to claim 1, characterized in that it further comprises a step of obtaining an information according to which the signal to be encoded is tonal or non-tonal and that the steps of calculating the threshold of masking and determination. of perceptual importance according to this masking threshold, are conducted only if the signal is non-tonal.

6. Method according to claim 1, characterized in that the improvement coding is a TDAC type enhancement coding in an extended coder whose core coding is of standard G.729.1 coder type.

A method of hierarchical decoding of a digital audio frequency signal into a plurality of frequency sub-bands comprising a core decoding of a received signal at a first rate and at least one higher rate improvement decoding of a residual signal, the heart decoding using a binary allocation according to an energy criterion, characterized in that it comprises the following steps for the improvement decoding:

 calculating a frequency masking threshold (605) for at least a portion of the frequency sub-bands processed by the enhancement decoding;

 determining (604) a perceptual importance per frequency subband as a function of the calculated masking threshold and as a function of the number of bits allocated for the core decoding;

 - bit allocation (604, 605) in the frequency sub-bands processed by the enhancement decoding, according to the determined perceptual importance; and

 -decoding (611) the residual signal according to the bit allocation.

8. Decoding method according to claim 7, characterized in that the step of determining a perceptual importance comprises:

 a first step (605) of defining a first perceptual importance for at least one frequency sub-band of the improvement decoding, as a function of the frequency masking threshold in the sub-band, of quantized values of the decoding of the spectral envelope for the frequency subband and a determined normalization factor;

a second step (604) of subtracting from the first perceptual importance a ratio between the number of bits allocated for the core decoding and the number of possible coefficients in said sub-band.

9. Hierarchical coder of a digital audio frequency input signal in a plurality of frequency sub-bands comprising an input signal core encoder at a first rate and at least one higher rate improvement encoder of a residual signal, the heart encoder using a binary allocation (506) according to an energy criterion, characterized in that the enhancement coder comprises:

 a module (511) for calculating a frequency masking threshold for at least part of the frequency bands processed by the improvement coder;

 a determination module (512) of perceptual importance by sub-frequency band according to the calculated masking threshold and as a function of the number of bits allocated for the core coder;

 a bit allocation module (512) of bits in the frequency sub-bands processed by the enhancement coder, according to the determined perceptual importance; and

 a module for coding the residual signal (513) according to the bit allocation.

10. Hierarchical decoder of a digital audio signal in several frequency subbands comprising a heart decoder of a signal received at a first rate and at least one higher rate of improvement decoder, a residual signal, the heart decoder. using a binary allocation according to an energy criterion, characterized in that it comprises the improvement decoder comprises:

a module for calculating a frequency masking threshold (605) for at least a portion of the frequency sub-bands processed by the enhancement decoder; a determination module (604) of perceptual importance per frequency subband as a function of the calculated masking threshold and as a function of the number of bits allocated for the heart decoder;

 a bit allocation module (604) in the frequency sub-bands processed by the enhancement decoder, according to the determined perceptual importance; and

 a decoding module (611) of the residual signal according to the bit allocation.

11. Computer program comprising code instructions for implementing the steps of a coding method according to one of claims 1 to

6, when executed by a processor.

Computer program comprising code instructions for implementing the steps of a decoding method according to one of claims 7 to 8, when executed by a processor.