EP2374124B1 - Advanced encoding of multi-channel digital audio signals - Google Patents

Description

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

More particularly, the present invention relates to the parametric encoding / decoding of multichannel audio signals.

This type of coding / decoding is based on the extraction of spatialization parameters so that at decoding, the spatial perception of the listener can be reconstituted.

Such a coding technique is known under the name of "Binaural Cue Coding" in English (BCC) which aims on the one hand to extract and then code the indices of auditory spatialization and on the other hand to code a monophonic or stereophonic signal from a mastering of the original multichannel signal.

This parametric approach is a low rate coding. The main advantage of this coding approach is to allow a better compression rate than conventional multi-channel digital audio compression methods while ensuring the backward compatibility of the compressed format obtained with the existing coding formats and broadcasting systems.

The MPEG Surround standard described in the MPEG document ISO / IEC 23003-1: 2007 and in the document " Breebaart, J. and Hotho, G. and Koppens, J. and Schuijers, E. and Oomen, W. and van de Par, S., "entitled" Background, concept, and architecture for the recent MPEG surround standard on multichannel audio Compression "in Journal of the Audio Engineering Society 55-5 (2007) 331-351 , describes a parametric coding structure as represented in figure 1 . Another example of a parametric coding structure is described in the Bin Cheng, Christian Ritz, Ian Burnett entitled "Encoding Independent Sources in Spatially Squeezed Surround Audio Coding" in Advances in Multimedia Information Processing -PCM2a7 Reading Notes in Computer Science Volume 4810, 2077, page 804-813 .

So, the figure 1 describes such a coding / decoding system in which the coder 100 constructs a sum signal ("downmix" in English) S _s by matrixing in 110 channels of the original multichannel signal S and provides via a parameter extraction module 120, a reduced set of parameters P which characterize the spatial content of the original multichannel signal.

At the decoder 150, the multichannel signal is reconstructed (S ') by a synthesis module 160 which takes into account both the sum signal and the transmitted parameters P.

The sum signal has a reduced number of channels. These channels can be encoded by a conventional audio encoder before transmission or storage. Typically, the sum signal has two channels and is compatible with conventional stereo broadcasting. Before transmission or storage, this sum signal can thus be encoded by any conventional stereo encoder. The signal thus coded is then compatible with the devices comprising the corresponding decoder which reconstruct the sum signal while ignoring the spatial data.

When this type of coding by matrixing a multichannel signal to obtain a sum signal, is carried out after transformation in the frequency space of the multichannel signal, problems of reconstruction of the multichannel signal can occur.

Indeed, in this case, there is not necessarily spatial coherence between the sum signal and the rendering system on which the signal can be reproduced. For example, when the sum signal contains two channels, a stereophonic reproduction must make it possible to respect the relative position of the sound sources in the reconstructed sound space. The left / right positioning of the sound sources must be able to be respected.

In addition, after frequency band mastering, the resulting sum signal is then transmitted to the decoder in the form of a time signal.

The transition from time-frequency space to time space involves interactions between frequency bands and near time frames that introduce troublesome artifacts and artifacts.

There is therefore a need for a parametric frequency band coding / decoding technique which makes it possible to limit the defects introduced by the passage of the time-frequency domain signals to the time domain and to control the spatial coherence between the multichannel audio signal and the sum signal resulting from a mastering of sound sources.

The present invention improves the situation.

• obtention de données représentatives de la direction des sources sonores de la scène sonore;
• sélection d'un ensemble de sources sonores de la scène sonore constituant des sources principales;
• adaptation des données représentatives de la direction des sources principales sélectionnées, en fonction de caractéristiques de restitution du signal multicanal, par modification de la position des sources pour obtenir un écartement minimum entre deux sources;
• détermination d'une matrice de mixage des sources principales en fonction des données adaptées;
• matriçage des sources principales par la matrice déterminée pour obtenir un signal somme avec un nombre réduit de canaux;
• codage des données représentatives de la direction des sources sonores et formation d'un flux binaire comportant les données codées, le flux binaire étant apte à être transmis parallèlement au signal somme.

For this purpose, it proposes a method of encoding a multichannel audio signal representing a sound scene comprising a plurality of sound sources. The method is such that it comprises a step of decomposing the multichannel signal into frequency bands and the following steps per frequency band:

• obtaining data representative of the direction of the sound sources of the sound scene;
• selecting a set of sound sources of the sound scene constituting main sources;
• adapting the data representative of the direction of the main sources selected, according to the multichannel signal reproduction characteristics, by modifying the position of the sources to obtain a minimum spacing between two sources;
• determination of a mixing matrix of the main sources according to the adapted data;
• mastering the main sources by the determined matrix to obtain a sum signal with a reduced number of channels;
• coding of the data representative of the direction of the sound sources and formation of a bit stream comprising the coded data, the bit stream being able to be transmitted parallel to the sum signal.

Thus, for obtaining the sum signal, the mixing matrix takes into account source direction information data. This makes it possible to adapt the resulting sum signal, for a good sound reproduction in the space during the reconstruction of this signal to the decoder. The sum signal is thus adapted to the multichannel signal reproduction characteristics and to the possible recoveries of the positions of the sound sources. The spatial coherence between the sum signal and the multichannel signal is thus respected.

The adaptation of the data modifying the position of the sources to obtain a minimum spacing between two sources thus makes it possible for the two sources that would be after sound reproduction too close to each other to be discarded so that the restitution of the signal makes it possible to the listener to differentiate the position of these sources.

By separately encoding the directional data and the sound sources per frequency band, it is exploited that the number of active sources in a frequency band is generally low, which increases the coding performance.

It is not necessary to transmit further reconstruction data from the mixing matrix to the decoder since this will be determined from the encoded direction data.

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 one embodiment, the data representative of the direction are directivity information representative of the distribution of the sound sources in the sound scene.

The directivity information associated with a source gives not only the direction of the source but also the shape, or spatial distribution, of the source, ie the interaction that this source can have with other sources of the sound stage.

Knowing this directivity information associated with the sum signal will allow the decoder to obtain a signal of better quality which takes into account interchannel redundancies in a global manner and the probable phase oppositions between channels.

In a particular embodiment, the coding of the directivity information is performed by a parametric representation method.

This method is of low complexity and adapts particularly to the case of synthetic sound scenes representing an ideal coding situation.

In another embodiment, the coding of the directivity information is performed by a principal component analysis method delivering basic directivity vectors associated with gains allowing the reconstruction of the initial directivities.

This thus makes it possible to code the directivities of complex sound scenes whose coding can not easily be represented by a model.

In yet another embodiment, the coding of the directivity information is performed by a combination of a principal component analysis method and a parametric representation method.

Thus, it is for example possible to perform the coding in parallel by the two methods and to choose the one that satisfies a criterion for optimizing the coding rate, for example.

It is also possible to perform these two methods in cascade so as to simply code a part of the directivities by the parametric coding method and for that which is not modeled, to carry out an encoding by the principal component analysis method. , so as to represent at best, all the directivities. The distribution of the flow between the two models of encoding directivités can be chosen according to a criterion of minimization of the error of reconstruction of the directivités.

In one embodiment of the invention, the method further includes encoding secondary sources among the unselected sources of the sound scene and inserting coding information of the secondary sources into the bit stream.

The coding of the secondary sources will thus make it possible to provide additional precision on the decoded signal, in particular for complex signals of the type, for example, ambiophonic ones.

• extraction dans le flux binaire et décodage de données représentatives de la direction des sources sonores dans la scène sonore;
• adaptation d'au moins une partie des données de direction en fonction de caractéristiques de restitution du signal multicanal, par modification de la position des sources obtenues par les données de direction, pour obtenir un écartement minimum entre deux sources;
• détermination d'une matrice de mixage du signal somme en fonction des données adaptées et calcul d'une matrice de mixage inverse;
• dématriçage du signal somme par la matrice de mixage inverse pour obtenir un ensemble de sources principales;
• reconstruction du signal audio multicanal par spatialisation au moins des sources principales avec les données extraites décodées.

The present invention also relates to a method for decoding a multichannel audio signal representing a sound scene comprising a plurality of sound sources, from a bit stream and a sum signal. The method is such that it comprises the following steps:

• extracting in the bit stream and decoding data representative of the direction of the sound sources in the sound scene;
• adapting at least a portion of the direction data according to the multichannel signal reproduction characteristics, by changing the position of the sources obtained by the direction data, to obtain a minimum spacing between two sources;
• determining a mixing matrix of the sum signal as a function of the adapted data and calculating an inverse mixing matrix;
• demapping the sum signal by the reverse mixing matrix to obtain a set of main sources;
• reconstruction of the multichannel audio signal by spatializing at least the main sources with the extracted data decoded.

The decoded direction data will thus make it possible to find the inverse mixing matrix of that used at the encoder. This mixing matrix makes it possible to find, from the sum signal, the main sources that will be rendered in space with a good spatial coherence.

The adaptation step thus makes it possible to find the directions of the sources to be spatialized so as to obtain a restitution of the sound which is coherent with the rendering system.

The reconstructed signal is then well adapted to the characteristics of restitution of the multichannel signal while avoiding possible recoveries of the positions of the sound sources.

Two sources too close are thus removed to be restored so that an auditor can differentiate them.

• extraction du flux binaire, d'informations de codage de sources secondaires codées;
• décodage des sources secondaires à partir des informations de codage extraites;
• regroupement des sources secondaires aux sources principales pour la spatialisation.

In one embodiment, the decoding method further comprises the following steps:

• extracting the bitstream, coding information from coded secondary sources;
• decoding the secondary sources from the extracted coding information;
• grouping of secondary sources to the main sources for spatialization.

The decoding of secondary sources then brings more precision to the sound stage.

• un module de décomposition du signal multicanal en bande de fréquence;
• un module d'obtention de données représentatives de la direction des sources sonores de la scène sonore;
• un module de sélection d'un ensemble de sources sonores de la scène sonore constituant des sources principales;
• un module d'adaptation des données représentatives de la direction des sources principales sélectionnées, en fonction de caractéristiques de restitution du signal multicanal, par des moyens de modification de la position des sources pour obtenir un écartement minimum entre deux sources;
• un module de détermination d'une matrice de mixage des sources principales en fonction des données issues du module d'adaptation;
• un module de matriçage des sources principales sélectionnées par la matrice déterminée pour obtenir un signal somme avec un nombre réduit de canaux;
• un module de codage des données représentatives de la direction des sources sonores; et
• un module de formation d'un flux binaire comportant les données codées, le flux binaire étant apte à être transmis parallèlement au signal somme.

The present invention also relates to an encoder of a multichannel audio signal representing a sound scene having a plurality of sound sources. The encoder is such that it comprises:

• a multichannel signal decomposition module in a frequency band;
• a module for obtaining data representative of the direction of the sound sources of the sound scene;
• a module for selecting a set of sound sources of the sound scene constituting main sources;
• a module for adapting the data representative of the direction of the main sources selected, as a function of the multichannel signal reproduction characteristics, by means of modifying the position of the sources to obtain a minimum spacing between two sources;
• a module for determining a mixing matrix of the main sources according to the data from the adaptation module;
• a mastering module of the main sources selected by the determined matrix to obtain a sum signal with a reduced number of channels;
• a coding module for data representative of the direction of the sound sources; and
• a module for forming a bit stream comprising the coded data, the bit stream being able to be transmitted parallel to the sum signal.

• un module d'extraction et de décodage de données représentatives de la direction des sources sonores dans la scène sonore;
• un module d'adaptation d'au moins une partie des données de direction en fonction de caractéristiques de restitution du signal multicanal, par des moyens de modification de la position des sources obtenues par les données de direction, pour obtenir un écartement minimum entre deux sources;
• un module de détermination d'une matrice de mixage du signal somme en fonction des données issues du module d'adaptation et de calcul d'une matrice de mixage inverse;
• un module de dématriçage du signal somme par la matrice de mixage inverse pour obtenir un ensemble de sources principales;
• un module de reconstruction du signal audio multicanal par spatialisation au moins des sources principales avec les données extraites décodées.

It also relates to a decoder of a multichannel audio signal representing a sound scene comprising a plurality of sound sources, receiving as input a bit stream and a sum signal. The decoder is such that it comprises:

• a module for extracting and decoding data representative of the direction of the sound sources in the sound scene;
• a module for adapting at least a portion of the direction data as a function of the multichannel signal reproduction characteristics, by means for modifying the position of the sources obtained by the direction data, to obtain a minimum spacing between two sources ;
• a module for determining a mixing matrix of the sum signal as a function of the data from the adaptation and calculation module of an inverse mixing matrix;
• a module for demapping the sum signal by the inverse mixing matrix to obtain a set of main sources;
• a module for reconstructing the multichannel audio signal by spatializing at least the main sources with the extracted decoded data.

Finally, it relates to a computer program comprising code instructions for implementing the steps of a coding method as described and / or a decoding method as described, when these instructions are executed by a processor. .

In a more general manner, a means of storage, readable by a computer or a processor, whether or not integrated into the encoder, possibly removable, stores a computer program implementing an encoding method and / or a decoding method according to the invention.

• la figure 1 illustre un système de codage/décodage de l'état de l'art de type système normalisé MPEG Surround;
• la figure 2 illustre un codeur et un procédé de codage selon un mode de réalisation de l'invention;
• la figure 3a illustre un premier mode de réalisation du codage des directivités selon l'invention;
• la figure 3b illustre un second mode de réalisation du codage des directivités selon l'invention;
• la figure 4 illustre un organigramme représentant les étapes de la détermination d'une matrice de mixage selon un mode de réalisation de l'invention;
• la figure 5a illustre un exemple de répartition de sources sonores autour d'un auditeur;
• la figure 5b illustre l'adaptation de la répartition de sources sonores autour d'un auditeur pour adapter les données de direction des sources sonores selon un mode de réalisation de l'invention;
• la figure 6 illustre un décodeur et un procédé de décodage selon un mode de réalisation de l'invention; et
• les figures 7a et 7b représentent respectivement un exemple de dispositif comprenant un codeur et un exemple de dispositif comprenant un décodeur selon l'invention.

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

• the figure 1 illustrates a state-of-the-art coding / decoding system of the standard MPEG Surround system type;
• the figure 2 illustrates an encoder and a coding method according to an embodiment of the invention;
• the figure 3a illustrates a first embodiment of the coding of the directivities according to the invention;
• the figure 3b illustrates a second embodiment of the coding of the directivities according to the invention;
• the figure 4 illustrates a flow chart showing the steps of determining a mixing matrix according to one embodiment of the invention;
• the figure 5a illustrates an example of distribution of sound sources around a listener;
• the figure 5b illustrates the adaptation of the distribution of sound sources around a listener to adapt the direction data of the sound sources according to one embodiment of the invention;
• the figure 6 illustrates a decoder and a decoding method according to an embodiment of the invention; and
• the Figures 7a and 7b respectively represent an example of a device comprising an encoder and an example of a device comprising a decoder according to the invention.

The figure 2 illustrates in the form of a block diagram, an encoder according to one embodiment of the invention and the steps of a coding method according to one embodiment of the invention.

All the processing in this encoder is performed by time frame. For reasons of simplification, the representation and description of the encoder as represented in FIG. figure 2 is done considering the processing carried out on a fixed time frame, without showing the temporal dependence in the notations.

The same treatment is however successively applied to all the time frames of the signal.

The encoder thus illustrated comprises a time-frequency transform module 210 which receives as input an original multichannel signal representing a sound scene comprising a plurality of sound sources.

This module therefore performs a step T of calculating the time-frequency transform of the original multichannel signal S _m . This transform is carried out for example by a short-term Fourier transform.

For this, each of the n _x channels of the original signal is window on the current time frame, then the Fourier transform F of the window signal is calculated using a fast calculation algorithm on n _FFT points. Thus, a complex X matrix of size n _FFT xn _x containing the coefficients of the original multichannel signal in the frequency space is obtained.

The subsequent processing by the encoder is done by frequency band. This is done by cutting the matrix of coefficients X into a set of sub-matrices X _j each containing the frequency coefficients in the j ^th band.

Different choices for the frequency division of the bands are possible. In order to ensure that the processing is applied to real signals, symmetrical bands with respect to the zero frequency in the Fourier transform are chosen in the short term. In addition, in order to optimize the coding efficiency, the choice of frequency bands approaching perceptual frequency scales, for example by choosing constant bandwidths in the ERB scales (for "equivalent Rectangular Bandwidth" in English) or Bark.

For the sake of simplification, the description of the coding steps performed by the coder will be made for a given frequency band. The steps are of course carried out for each of the frequency bands to be processed.

At the output of the module 210, the signal is thus obtained for a given frequency band S _fj .

A module for obtaining data of directions of the sound sources 220 makes it possible to determine, by a step OBT, on the one hand, the direction data associated with each of the sources of the sound stage and, on the other hand, to determine the sources of the sound stage for the given frequency band.

The direction data can be for example arrival direction data of a source that corresponds to the position of the source.

Data of this type are for example described in the document by Mr. Goodwin, JM. Jot, "Analysis and synthesis for universal spatial audio coding", 121 ^st AES Convention, October 2006.

In another embodiment, the direction data is intensity difference data between the sound sources. These differences in intensity make it possible to define average positions of the sources. They take for example the CLD (for "Channel Level Differences" in English) for the standard encoder MPEG Surround.

In the embodiment described here in more detail, the data representative of the directions of the sources are directional information.

The directivity information is representative of the spatial distribution of the sound sources in the sound scene.

The directivities are vectors of the same dimension as the number n _s of channels of the multichannel signal S _m .

Each source is associated with a vector of directivity.

For a multichannel signal, the directivity vector associated with a source corresponds to the weighting function to be applied to this source before playing it on a loudspeaker, so as to reproduce at best a direction of arrival and a width of source.

It is easily understood that for a very large number of regularly spaced loudspeakers, the directivity vector makes it possible to faithfully represent the radiation of a sound source. In the presence of an ambiophonic signal, the vector of directivity is obtained by the application of an inverse spherical Fourier transform on the components of the ambiophonic orders. Indeed, the ambiophonic signals correspond to a decomposition into spherical harmonics, hence the direct correspondence with the directivity of the sources.

The set of directivity vectors therefore constitutes a large amount of data that would be too expensive to transmit directly for low coding rate applications. To reduce the amount of information to be transmitted, two methods of representing the directivities can for example be used.

The Cod.Di coding module 230 of the directional information can thus implement one of the two methods described below or a combination of the two methods.

A first method is a parametric modeling method that makes it possible to exploit knowledge a priori on the signal format used. It consists of transmitting only a very small number of parameters and reconstructing the directivities according to known coding schemes.

For example, it is a question of exploiting the knowledge on the coding of the plane waves for signals of type ambiophonique to transmit only the value of the direction (azimuth and elevation) of the source. With this information, it is then possible to reconstruct the directivity corresponding to a plane wave coming from this direction.

For example, for a defined surround order, the associated directivity is known as a function of the direction of arrival of the sound source. There are several methods for estimating model parameters. Thus a search for peaks in the directivity diagram (by analogy with the sinusoidal analysis, as explained for example in the document " Computer modeling called musical sound (analysis, transformation, synthesis) "by Sylvain Marchand, PhD thesis, Bordeaux University 1 , allows to detect relatively accurately the direction of arrival.

Other methods such as "matching pursuit", as presented in S. Mallat, Z. Zhang, Matching pursuit with time-frequency dictionaries, IEEE Transactions on Signal Processing 41 (1993) 3397-3415 , or parametric spectral analysis can also be used in this context.

A parametric representation can also use a simple form dictionary to represent the directivities. During the coding of the directivities, one associates with an element of the dictionary, a datum for example the corresponding azimuth and a gain allowing to play on the amplitude of this vector of directivity of the dictionary. It is thus possible, from a dictionary of directivity form, to deduce the best form or the combination of forms that will best reconstruct the initial directivity.

For the implementation of this first method, the directivity coding module 230 comprises a parametric modeling module which outputs P directionality parameters. These parameters are then quantized by the quantization module 240.

This first method makes it possible to obtain a very good level of compression when the scene corresponds to an ideal coding. This will be particularly the case on synthetic soundtracks.

However, for complex scenes or microphonic sound, it is necessary to use more generic coding models, involving the transmission of a greater amount of information.

The second method described below makes it possible to overcome this disadvantage. In this second method, the representation of the directivity information is in the form of a linear combination of a limited number of basic directivities. This method is based on the fact that the set of directivities at a given moment generally has a reduced dimension. Indeed, only a small number of sources is active at a given moment and the directivity for each source varies little with the frequency.

It is thus possible to represent all the directivities in a group of frequency bands from a very small number of well-chosen basic directivities. The parameters transmitted are then the basic directivity vectors for the group of bands considered, and for each directivity to be coded, the coefficients to be applied to the basic directivities to reconstruct the directivity considered.

This method is based on a principal component analysis (PCA or PCA) method. This tool is widely developed by IT Jolliffe in "Principal Component Analysis", Springer, 2002 . The application of the principal component analysis to the coding of the directivities is carried out as follows: firstly, a matrix of the initial directivities Di is formed, whose number of lines corresponds to the total number of sources of the scene sound, and the number of columns corresponds to the number of channels of the original multichannel signal. Then, one carries out properly the principal component analysis which corresponds to the diagonalization of the covariance matrix, and which gives the matrix of the eigenvectors. Finally, we select the eigenvectors carrying the most important information and corresponding to the eigenvalues of higher value. The number of eigenvectors to keep may be fixed or variable in time depending on the available flow. This new base thus gives the matrix D _B ^T. The gain coefficients associated with this base are easily calculated from G _D = Di.D _B ^T.

In this embodiment, the representation of the directivities is therefore done from basic directivities. The matrix of directivities Di is written as the linear combination of these basic directivities. Thus one can write Di = G _D D _B , where D _B is the matrix of the basic directivities for all the bands and G _D the matrix of the associated gains. The number of rows of this matrix represents the total number of sources of the sound stage and the number of columns represents the number of basic directivity vectors.

In a variant of this embodiment, basic directivities are sent by group of considered bands, in order to more accurately represent the directivities. For example, it is possible to provide two directivity groups of base: one for low frequencies and one for high frequencies. The limit between these two groups can for example be chosen between 5 and 7 kHz.

For each frequency band, the vector of gain associated with the basic directivities is thus transmitted.

For this embodiment, the coding module 230 comprises a main component analysis module delivering basic directivity vectors D _B and associated coefficients or gain vectors G _D.

Thus, after PCR, a limited number of directivity vectors will be encoded and transmitted. For this, we use a scalar quantization performed by the quantization module 240, coefficients and vectors of basic directivities. The number of basic vectors to be transmitted may be fixed, or else selected by the coder by using, for example, a threshold on the mean square error between the original directivity and the reconstructed directivity. Thus, if the error is below the threshold, the base vector (s) hitherto selected (s) are sufficient, it is then not necessary to code an additional base vector.

In alternative embodiments, the coding of the directivities is achieved by a combination of the two representations listed above. The figure 3a illustrates in detail, the direction coding block 230, in a first embodiment.

This coding mode uses the two diagrams of representation of the directivities. Thus, a module 310 performs parametric modeling as previously explained to provide directional parameters (P).

A module 320 performs principal component analysis to provide both basic directivity vectors (D _B ) and associated coefficients (G _D ).

In this variant, a selection module 330 selects frequency band per frequency band, the best mode of coding for the directivity by choosing the best compromise reconstruction of the directivities / flow.

For each directivity, the choice of the representation chosen (parametric representation or by linear combination of basic directivities) is done in order to optimize the efficiency of the compression.

A selection criterion is, for example, the minimization of the mean squared error. A perceptual weighting may possibly be used for the choice of the directivity coding mode. This weighting is intended for example to promote the reconstruction of the directivities in the frontal area, for which the ear is more sensitive. In this case, the error function to be minimized in the case of the ACP encoding model can be in the following form: $$\mathrm{E}\mathrm{=}{\left(\mathrm{W}\left(\mathrm{di}\mathrm{-}{\mathrm{BOY\; WUT}}_{\mathrm{D}}{\mathrm{D}}_{\mathrm{B}}\right)\right)}^{\mathrm{2}}$$

With Di, the original directivities and W, the perceptual weighting function.

The directivity parameters from the selection module are then quantized by the quantization module 240 of the figure 2 .

In a second variant of the coding block 230, the two coding modes are cascaded. The figure 3b illustrates in detail this block of coding. Thus, in this variant embodiment, a parametric modeling module 340 performs a modeling for a certain number of directivities and outputs at the same time directivity parameters (P) for the modeled directivities and unmodelled directivities or residual directivities DiR .

These residual directivities (DiR) are encoded by a main component analysis module 350 which outputs basic directional vectors (D _B ) and associated coefficients (G _D ).

The directivity parameters, the basic directivity vectors as well as the coefficients are provided at the input of the quantization module 240 of the figure 2 .

Quantization Q is performed by reducing the accuracy as a function of perception data and then applying entropy coding. Also, the possibility of exploiting the redundancy between frequency bands or between successive frames can reduce the flow. Intra-frame or inter-frame predictions on the parameters can therefore be used. In general, the standard methods of quantification can be used. On the other hand, the vectors to be quantified being orthonormed, this property can be exploited during the scalar quantization of the components of the vector. Indeed, for a N-dimensional vector, only N-1 components will have to be quantified, the last component can be recalculated.

At the output of the coding module 230, the data of directions Di of the figure 2 , the parameters thus intended for the decoder are decoded by the internal decoding module 235 to find the same information that the decoder will have after receiving the coded direction data for the main sources selected by the module 260 described later. We thus obtain principal directions.

In the case of direction data in the form of direction of arrival of the sources, the information can be taken into account as it is.

When the data is in the form of difference in intensity between the sources, a step of calculating the average position of the sources is performed to use this information in the module for determining the mixing matrix 275.

Finally, when the data is directivity information, the module 235 determines a unique position by source by averaging the directivities. This average can for example be calculated as the barycenter of the directivity vector. These unique positions or main directions are then used by module 275.

This first determines the directions of the main sources and adapts them according to spatial coherence criterion, knowing the multichannel signal reproduction system.

In the case of stereophonic reproduction for example, the restitution takes place by two loudspeakers located at the front of the listener.

In this case, the steps implemented by the module 275 are described with reference to the figure 4 .

Thus, from the information on the position of the sources as well as the knowledge of the rendering characteristics, the sources positioned at the rear of the listener are brought forward to the step E30 of the figure 4 .

With reference to Figures 5a and 5b , the stages of adaptation of the position of the sources are illustrated. So, the figure 5a represents an original sound scene with 4 sound sources (A, B, C and D) distributed around the listener.

Sources C and D are located behind the listener centered in the center of the circle. Sources C and D are brought to the front of the stage by symmetry.

The figure 5b illustrates in the form of arrows, this operation.

Step E 31 of the figure 4 performs a test to see if the previous operation causes a recovery of the positions of the sources in the space. In the example of the figure 5b this is for example the case for sources B and D which after the operation of step E30, are located at a distance that does not differentiate them.

If there are sources in such a situation (positive test of step E31), step E32 modifies the position of one of the two sources in question to position it at a minimum distance e _min that allows the listener to differentiate these interlocutors. The spacing is symmetrically with respect to the equidistant point of the two sources to minimize the displacement of each. If the sources are placed too close to the limit of the sound image (extreme left or right), we position the source closest to this limit to this limit position, and place the other source with the minimum spacing by report to the first source.

In the example shown in figure 5b , it is the source B which is shifted so that the distance e _min separates the sources B and D.

If the test of step E31 is negative, the positions of the sources are maintained and step E33 is implemented. This step consists of constructing a mixing matrix from the source position information thus defined in the previous steps.

In the case of a signal reproduction by a 5.1 type system, the speakers are distributed around the listener. It is then not necessary to implement step E30 which brings the sources located at the rear of the listener forward.

On the other hand, step E32 of modifying the distances between two sources is possible. Indeed, when you want to position a sound source between two speakers of the 5.1 playback system, two sources may be located at a distance that does not allow the listener to differentiate them.

The directions of the sources are therefore modified to obtain a minimum distance between two sources, as explained above.

The mixing matrix is thus determined in step E33, as a function of the directions obtained after or without modifications.

This matrix is constructed in such a way as to ensure the spatial coherence of the sum signal, ie if it is rendered alone, the sum signal already makes it possible to obtain a sound scene where the relative position of the sound sources is respected: a frontal source in the original scene will be well perceived in front of the listener, a source on the left will be seen on the left, a source on the left will also be perceived more to the left, likewise on the right.

With these new angle values, an invertible matrix is built.

The different variants of mixing matrix choices are related to the different laws of spatial distribution or "panning" in English (sinus law, tangent, etc ...) presented in " Spatial sound generation and perception by amplitude panning techniques ", PhD thesis, Helsinki University of Technology, Espoo, Finland, 2001, V. Pulkki .

One can, for example, advantageously choose to represent the right lanes by a sine form and the left lanes by a cosine form, so as to make this matrix reversible.

On the other hand, for the extreme positions (-45 ° and 45 °) to be well represented, we can for example choose weighting coefficients set to 1 for the left channel and 0 for the right channel to represent the signal at the -45 ° position and vice versa to represent the 45 ° signal.

For the central position, at 0 °, to be well represented, the matrixing coefficients for the left channel and the right channel must be equal.

An example of determining the mixing matrix is explained below.

$${\mathrm{g}}_{\mathrm{Ds}\mathrm{2}}\mathrm{=}{\mathrm{sin\theta }}_{\mathrm{s}\mathrm{1}}$$

By choosing the panning law as a tangent law, the gains associated with a source for a stereophonic sum signal (2 channels) are calculated as follows: $${\mathrm{boy\; Wut}}_{\mathrm{gs}\mathrm{1}}\mathrm{=}{\cos }_{\mathrm{s}\mathrm{1}}$$

$${\mathrm{boy\; Wut}}_{\mathrm{ds}\mathrm{2}}\mathrm{=}{\mathrm{sin\theta }}_{\mathrm{s}\mathrm{1}}$$

θ _S1 , being the angle between the source 1 and the left speaker, considering the opening between the loudspeakers of 90 °.

$$\mathrm{Avec}M=\left[\begin{array}{cc}{g}_{\mathit{Gs}1}& g_{\mathit{Ds}1}\\ g_{\mathit{Gs}2}& g_{\mathit{Ds}2}\end{array}\right]$$

The sum signal S _sfi is thus obtained by the following operation: $${\mathrm{S}}_{\mathrm{ifc}}\mathrm{=}{\mathrm{S}}_{\mathrm{princ}}\mathrm{M}$$

$$\mathrm{With}M=\left[\begin{array}{cc}{\mathrm{boy\; Wut}}_{\mathit{gs}1}& {\mathrm{boy\; Wut}}_{\mathit{ds}1}\\ {\mathrm{boy\; Wut}}_{\mathit{gs}2}& {\mathrm{boy\; Wut}}_{\mathit{ds}2}\end{array}\right]$$

Going back to the description of the figure 2 , the encoder as described herein further comprises a selection module 260 adapted to select in step Select main sources (S _princ ) among the sources of the sound scene to be encoded (S _tot ).

For this, a particular embodiment uses a principal component analysis method, ACP, in each frequency band in block 220 to extract all the sources of the sound scene (S _tot ). This analysis makes it possible to classify the sources in subbands in order of importance according to the level of energy for example.

The sources of greater importance (therefore of greater energy) are then selected by the module 260 to constitute the main sources (S _princ ), which are then stamped by the module 270, by the matrix M as defined by the module 275 , to build a sum signal (S _sfi ) (or "downmix" in English).

This sum signal per frequency band undergoes an inverse time-frequency transform T ^-1 by the inverse transform module 290 to provide a time sum signal (S _s ). This sum signal is then encoded by a speech coder or an audio coder of the state of the art (for example: G.729.1 or MPEG-4 AAC).

Secondary sources (S _sec ) may be encoded by a coding module 280 and added to the bitstream in the bitstream building module 250.

For these secondary sources, that is to say the sources that are not transmitted directly in the sum signal, there are different alternatives of treatments.

Since these sources are considered non-essential to the sound stage, they may not be transmitted.

However, it is possible to code some or all of these secondary sources by the coding module 280 which may in one embodiment be a short-term Fourier transform coding module. These sources can then be separately encoded using the aforementioned audio or speech coders.

In a variant of this coding, the coefficients of the transform of these secondary sources can be coded directly only in the bands considered to be important.

The secondary sources can be encoded by parametric representations, these representations can be in the form of spectral envelope or temporal envelope.

These representations are coded in the step Cod.S _sec of the module 280 and inserted in the step Con.Fb of the module 250, in the bit stream with the quantized coded directivity information. These parametric representations then constitute coding information of the secondary sources.

In the case of certain multichannel signals, in particular of the ambiophonic type, the encoder as described implements an additional preprocessing step P by a pre-processing module 215.

This module performs a base change step in order to express the sound scene using the plane wave decomposition of the acoustic field.

The original surround signal is seen as the angular Fourier transform of a sound field. Thus the different components represent the values for the different angular frequencies. The first plane wave decomposition operation therefore corresponds to taking the omnidirectional component of the ambiophonic signal as representing the zero angular frequency (this component is therefore a real component). Then, The following surround components (order 1, 2, 3, etc.) are combined to obtain the complex coefficients of the angular Fourier transform.

For a more precise description of the ambiophonic format, we can refer to the thesis of Jérôme Daniel, entitled "Representation of acoustic fields, application to the transmission and reproduction of complex sound scenes in a multimedia context" 2001, Paris 6 .

Thus, for each surround order greater than 1 (in 2-dimensions), the first component represents the real part, and the second component represents the imaginary part. For a two-dimensional representation, for an order O, we obtain O + 1 complex components. A Short Term Fourier Transform (on the time dimension) is then applied to obtain the Fourier transforms (in the frequency domain) of each angular harmonic. This step then integrates the transformation step T of the module 210. the complete angular transform by recreating the harmonics of negative frequencies by Hermitian symmetry. Finally, an inverse Fourier transform is carried out on the dimension of the angular frequencies to pass in the domain of the directivities.

This preprocessing step P allows the coder to work in a signal space whose physical and perceptual interpretation is simplified, which makes it possible to more effectively exploit the knowledge of spatial auditory perception and thus to improve the coding performances. The encoding of the surround signals, however, remains possible without this pre-processing step.

For non-surround signals, this step is not necessary. For these signals, the knowledge of the recording system or reproduction associated with the signal makes it possible to directly interpret the signals as a plane wave decomposition of the acoustic field.

The figure 6 describes now a decoder and a decoding method in an embodiment of the invention.

This decoder receives as input the bit stream F _b as constructed by the encoder described above as well as the sum signal S _s .

In the same way as for the encoder, all the processing is done by time frame. To simplify the notations, the description of the decoder which follows only describes the processing performed on a fixed time frame and does not show the temporal dependence in the notations. In the decoder, however, this same processing is successively applied to all the time frames of the signal.

The decoder thus described comprises a decoding module 650 Decod.Fb information contained in the bit stream Fb received.

The direction information and more particularly here, directivités are extracted from the bit stream.

The possible outputs of this decoding module of the bitstream depend on the coding methods of the directivities used in the coding. They can be in the form of basic directivity vectors D _B and associated coefficients G _D and / or modeling parameters P.

This data is then transmitted to a directional information reconstruction module 660 which performs the decoding of the directivity information by reverse operations from those performed in the coding.

The number of directivity to be reconstructed is equal to the number n _tot of sources in the frequency band considered, each source being associated with a directivity vector.

In the case of the representation of directivities from basic directivity, the matrix of directivities Di is written as the linear combination of these basic directivities. Thus one can write Di = G _D D _B , where D _B is the matrix of the basic directivities for all the bands and G _D the matrix of the associated gains. This gain matrix has a number of lines equal to the total number of sources n _tot , and a number of columns equal to the number of basic directivity vectors.

In a variant of this embodiment, basic directivities are decoded by group of frequency bands considered, in order to more accurately represent the directivities. As explained for coding, one can for example provide two groups of basic directivities: one for low frequencies and one for high frequencies. A vector of gains associated with the basic directivities is then decoded for each band.

In the end we reconstruct as many directivities as sources. These directivities are grouped in a matrix Di where the lines correspond to the angle values (as much angle value as channels in the multichannel signal to be reconstructed), and each column corresponds to the directivity of the corresponding source, it is to say that the column r of Di gives the directivity of the source which is in the column r of S.

A module 690 for defining the main directions of the sources and for determining the mixing matrix N receives these computerized directions or decoded directivities.

This module first calculates the main directions by, for example, averaging the directivities received to find the directions. In function of these directions, a mixing matrix, inverse to that used for the coding is determined.

Knowing the panning laws used for the mixing matrix at the encoder, the decoder is able to reconstruct the inverse mixing matrix with the directions information corresponding to the directions of the main sources.

The directivity information is transmitted separately for each source. Thus, in the bitstream, the directivities relative to the main sources and the directivities of the secondary sources are well identified.

It should be noted that this decoder does not need any other information to calculate this matrix since it depends on the direction information received in the bit stream.

The same algorithm as that described with reference to the figure 4 is then implemented in the module 690 to find the mixing matrix adapted to the restitution provided for the sum signal.

The number of rows of the matrix N corresponds to the number of channels of the sum signal, and the number of columns corresponds to the number of main sources transmitted.

The inverse matrix N as defined is then used by the demosaicing module 620.

The decoder thus receives in parallel the bit stream, the sum signal S _s . It undergoes a first time-frequency transform step T by the transform module 610 to obtain a sum signal per frequency band. S _sfi

This transform is carried out using, for example, the short-term Fourier transform. It should be noted that other transforms or filterbanks may also be used, including non-uniform filterbanks according to a perception scale (e.g. Bark). It may be noted that in order to avoid discontinuities during the reconstruction of the signal from this transform, a recovery addition method is used.

For the time frame considered, the step of calculating the Fourier transform in the short term consists of winding each of the n _f channels of the sum signal S _s using a window w of length greater than the time frame, then to compute the Fourier transform of the window signal using a fast computation algorithm on N _FFT points. Size of a complex matrix F is thus obtained n _FFT xn _f containing the coefficients of the sum signal in the frequency space.

In the following, the entire processing is done in frequency bands. For this, the matrix coefficients F is cut into a plurality of submatrices F _j each containing the frequency coefficients in the j ^th band. Different choices for the frequency division of the bands are possible. In order to ensure that the processing is applied to real signals, symmetrical bands with respect to the zero frequency in the Fourier transform are chosen in the short term. In addition, in order to optimize the decoding efficiency, preference is given to the choice of frequency bands approaching perceptual frequency scales, for example by choosing constant bandwidths in the ERB or Bark scales.

For reasons of simplification, the description of the decoding steps performed by the decoder will be made for a given frequency band. The steps are of course carried out for each of the frequency bands to be processed.

The frequency coefficients of the signal transform sum of the frequency band considered are stamped by the module 620 by the matrix N determined according to the determination step described above so as to find the main sources of the sound scene.

• S_princ=BN, où N est de dimension n_f x n_princ et B est une matrice de dimension n_binx n_f où n_bin est le nombre de composantes (ou bins) fréquentielles retenues dans la bande de fréquence considérée.

More specifically, the S _princ matrix of frequency coefficients for the current frequency band of the n _princ main sources is obtained according to the relation:

• S _princ = BN, where N is of dimension n _f xn _princ and B is a matrix of dimension n _bin xn _f where n _bin is the number of components (or bins) frequency retained in the considered frequency band.

The lines of B are the frequency components in the current frequency band, the columns correspond to the channels of the sum signal. The lines of S _princ are the frequency components in the current frequency band, and each column corresponds to a main source.

When the scene is complex, it may happen that the number of sources to be reconstructed in the current frequency band to obtain a satisfactory reconstruction of the scene is greater than the number of channels of the sum signal.

In this case, additional or secondary sources are coded and then decoded from the bitstream for the current band by the module 650 for decoding the bitstream.

This decoding module then decodes, in addition to the directional information, the secondary sources.

The decoding of the secondary sources is carried out by the inverse operations that those which were carried out with the coding.

Whatever the coding method that has been chosen for the secondary sources, if secondary source reconstruction data has been transmitted in the bit stream for the current band, the corresponding data are decoded to reconstruct the _dry matrix S of the frequency coefficients in the current band of the n _sec secondary sources. The shape of the _dry matrix S is similar to the matrix S _princ , that is, the lines are the frequency components in the current frequency band, and each column corresponds to a secondary source.

It is thus possible to construct the complete matrix S at 680, frequency coefficients of the set of n _tot = n _princ + n _sec sources necessary for the reconstruction of the multichannel signal in the band considered, obtained by combining the two matrices S _princ and S _supp according to the relation S = ( S _princ S _stupp ) . S is therefore a matrix of dimension n _bin xn _tot . Also, the form is identical to the matrices S _princ and S _supp : the lines are the frequency components in the current frequency band, each column is a source, with n _tot sources in total.

• Y=SD^T, où Y est le signal reconstruit dans la bande. Les lignes de la matrice Y sont les composantes fréquentielles dans la bande de fréquence courante, et chaque colonne correspond à un canal du signal multicanal à reconstruire.

From the matrix S of the coefficients of the sources and of the matrix Di of the associated directivities, the frequency coefficients of the multichannel signal reconstructed in the band are calculated in the spatialization module 630, according to the relation:

• Y = SD ^T , where Y is the reconstructed signal in the band. The rows of the matrix Y are the frequency components in the current frequency band, and each column corresponds to a channel of the multichannel signal to be reconstructed.

By reproducing the same processing in each of the frequency bands, the complete Fourier transforms of the signal channels to be reconstructed for the current time frame are reconstructed. The corresponding time signals are then obtained by inverse Fourier transform T ^-1 , using a fast algorithm implemented by the inverse transform module 640.

This gives the multichannel signal S _m on the current time frame. The different time frames are then combined by conventional overlap-add (or overlap-add) method to reconstruct the complete multichannel signal.

In general, temporal or frequency smoothing of the parameters can be used both for analysis and synthesis to ensure smooth transitions in the sound scene. A sign of sudden change of the sound stage may be reserved in the bit stream to avoid smoothing the decoder in the case of detection of a rapid change in the composition of the sound stage. On the other hand, conventional methods of adapting the resolution of the time-frequency analysis can be used (change in the size of the analysis and synthesis windows over time).

In the same way as the encoder, a base change module can perform a pre-processing to obtain a plane wave decomposition of the signals, a base change module 670 performs the inverse operation P ^-1 from the signals. in plane waves to find the original multichannel signal.

Encoders and decoders as described with reference to figures 2 and 6 can be integrated in a multimedia equipment type decoder lounge, computer or communication equipment such as a mobile phone or personal electronic diary.

The figure 7a represents an example of such a multimedia equipment or coding device comprising an encoder according to the invention. This device comprises a PROC processor cooperating with a memory block BM having a storage and / or working memory MEM.

The device comprises an input module adapted to receive a multichannel signal representing a sound scene, either by a communication network, or by reading a content stored on a storage medium. This multimedia equipment may also include means for capturing such a multichannel signal.

• obtention de données représentatives de la direction des sources sonores de la scène sonore;
• sélection d'un ensemble de sources sonores de la scène sonore constituant des sources principales;
• adaptation des données représentatives de la direction des sources principales sélectionnées, en fonction de caractéristiques de restitution du signal multicanal;
• détermination d'une matrice de mixage des sources principales en fonction des données adaptées;
• matriçage des sources principales par la matrice déterminée pour obtenir un signal somme avec un nombre réduit de canaux:
• codage des données représentatives de la direction des sources sonores et formation d'un flux binaire comportant les données codées, le flux binaire étant apte à être transmis parallèlement au signal somme.

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 decomposition of the multichannel signal in frequency bands and the following steps per frequency band:

• obtaining data representative of the direction of the sound sources of the sound scene;
• selecting a set of sound sources of the sound scene constituting main sources;
• adapting the data representative of the direction of the main sources selected, according to the multichannel signal reproduction characteristics;
• determination of a mixing matrix of the main sources according to the adapted data;
• mastering the main sources by the determined matrix to obtain a sum signal with a reduced number of channels:
• coding of the data representative of the direction of the sound sources and formation of a bit stream comprising the coded data, the bit stream being able to be transmitted parallel to the sum signal.

Typically, the description of the figure 2 takes 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 device or downloadable in the memory space of the equipment.

The device comprises an output module capable of transmitting a bit stream Fb and a sum signal Ss resulting from the coding of the multichannel signal.

In the same way, figure 7b illustrates an example of multimedia equipment or decoding device comprising a decoder according to the invention.

This device comprises a PROC processor cooperating with a memory block BM having a storage and / or working memory MEM.

The device comprises an input module adapted to receive a bit stream Fb and a sum signal S _s coming for example from a communication network. These input signals can come from a reading on a storage medium.

• d'adaptation d'au moins une partie des données de direction en fonction de caractéristiques de restitution du signal multicanal;
• de détermination d'une matrice de mixage du signal somme en fonction des données adaptées et de calcul d'une matrice de mixage inverse;
• de dématriçage du signal somme par la matrice de mixage inverse pour obtenir un ensemble de sources principales;
• de reconstruction du signal audio multicanal par spatialisation au moins des sources principales avec les données extraites décodées.

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 extraction in the bitstream and decoding of data representative of the direction of the sound sources in the sound scene;

• adapting at least a portion of the direction data according to characteristics of rendering of the multichannel signal;
• determining a mixing matrix of the sum signal according to the adapted data and calculating an inverse mixing matrix;
• demapping the sum signal by the reverse mixing matrix to obtain a set of main sources;
• for reconstructing the multichannel audio signal by spatializing at least the main sources with the decoded extracted data.

Typically, the description of the figure 6 takes 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 device or downloadable in the memory space of the equipment.

The device comprises an output module capable of transmitting a multichannel signal decoded by the decoding method implemented by the equipment.

This multimedia equipment may also include speaker-type reproduction means or communication means capable of transmitting this multi-channel signal.

Obviously, such multimedia equipment may include both the encoder and the decoder according to the invention. The input signal then being the original multichannel signal and the output signal, the decoded multichannel signal.

Claims (11)

1. Method for coding a multi-channel audio signal representing a sound scene comprising a plurality of sound sources, characterized in that it comprises a step (T) of decomposing the multi-channel signal into frequency bands and the following steps per frequency band:

   - obtaining (OBT) of data representative of the direction of the sound sources of the sound scene;

   - selection (Select) of a set of sound sources of the sound scene constituting principal sources;

   - adaptation (DiA_M) of the data representative of the direction of the selected principal sources, as a function of restitution characteristics of the multi-channel signal, by modification of the position of the sources so as to obtain a minimum separation between two sources;

   - determination (DiA_M) of a matrix for mixing the principal sources as a function of the adapted data;

   - matrixing (M) of the principal sources by the matrix determined so as to obtain a sum signal with a reduced number of channels;

   - coding (Cod.Di) of the data representative of the direction of the sound sources and formation of a binary stream comprising the coded data, the binary stream being able to be transmitted in parallel with the sum signal.
2. Method according to Claim 1, characterized in that the data representative of the direction are information regarding directivities representative of the distribution of the sound sources in the sound scene.
3. Method according to Claim 2, characterized in that the coding of the information regarding directivities is performed by a parametric representation procedure.
4. Method according to Claim 2, characterized in that the coding of the directivity information is performed by a principal component analysis procedure delivering base directivity vectors associated with gains allowing the reconstruction of the initial directivities.
5. Method according to Claim 2, characterized in that the coding of the directivity information is performed by a combination of a principal component analysis procedure and of a parametric representation procedure.
6. Method according to Claim 1, characterized in that it furthermore comprises the coding of secondary sources from among the unselected sources of the sound scene and insertion of coding information for the secondary sources into the binary stream.
7. Method for decoding a multi-channel audio signal representing a sound scene comprising a plurality of sound sources, with the help of a binary stream and of a sum signal, characterized in that it comprises the following steps:

   - extraction (Decod. Fb) from the binary stream and decoding of data representative of the direction of the sound sources in the sound scene;

   - adaptation (DiA_N) of at least some of the direction data as a function of restitution characteristics of the multi-channel signal, by modification of the position of the sources obtained by the direction data, so as to obtain a minimum separation between two sources;

   - determination (DiA_N) of a matrix for mixing the sum signal as a function of the adapted data and calculation of an inverse mixing matrix;

   - dematrixing (N) of the sum signal by the inverse mixing matrix so as to obtain a set of principal sources;

   - reconstruction (SPAT.) of the multi-channel audio signal by spatialization at least of the principal sources with the decoded extracted data.
8. Decoding method according to Claim 7, characterized in that it furthermore comprises the following steps:

   - extraction, from the binary stream, of coding information for coded secondary sources;

   - decoding of the secondary sources with the help of the coding information extracted;

   - grouping of the secondary sources with the principal sources for the spatialization.
9. Coder of a multi-channel audio signal representing a sound scene comprising a plurality of sound sources, characterized in that it comprises:

   - a module (210) for decomposing the multi-channel signal into frequency bands;

   - a module (220) for obtaining data representative of the direction of the sound sources of the sound scene;

   - a module (260) for selecting a set of sound sources of the sound scene constituting principal sources;

   - a module (275) for adapting the data representative of the direction of the selected principal sources, as a function of restitution characteristics of the multi-channel signal, by means for modifying the position of the sources so as to obtain a minimum separation between two sources;

   - a module (275) for determining a matrix for mixing the principal sources as a function of the data arising from the adaptation module;

   - a module (270) for matrixing the principal sources selected by the matrix determined so as to obtain a sum signal with a reduced number of channels;

   - a module (230) for coding the data representative of the direction of the sound sources; and

   - a module (250) for forming a binary stream comprising the coded data, the binary stream being able to be transmitted in parallel with the sum signal.
10. Decoder of a multi-channel audio signal representing a sound scene comprising a plurality of sound sources, receiving as input a binary stream and a sum signal, characterized in that it comprises:

    - a module (650) for extracting and decoding data representative of the direction of the sound sources in the sound scene;

    - a module (690) for adapting at least some of the direction data as a function of restitution characteristics of the multi-channel signal, by means for modifying the position of the sources obtained by the direction data, so as to obtain a minimum separation between two sources;

    - a module (690) for determining a matrix for mixing the sum signal as a function of the data arising from the module for adapting and for calculating an inverse mixing matrix;

    - a module (620) for dematrixing the sum signal by the inverse mixing matrix so as to obtain a set of principal sources;

    - a module (630) for reconstructing the multi-channel audio signal by spatialization at least of the principal sources with the decoded extracted data.
11. Computer program comprising code instructions for the implementation of the steps of a coding method according to one of Claims 1 to 6 and/or of a decoding method according to either of Claims 7 and 8, when these instructions are executed by a processor.

Images