EP3400599B1 - Improved ambisonic encoder for a sound source having a plurality of reflections - Google Patents

Description

FIELD OF THE INVENTION

The present invention relates to the ambisonic encoding of sound sources. It relates more specifically to improving the efficiency of this coding, in the case where a sound source is affected by reflections in a sound scene.

PREVIOUS STATE OF THE ART

Spatialized representations of sound bring together techniques for synthesizing and reproducing a sound environment, allowing the listener to be much more immersed in a sound environment. They allow a user in particular to discern a number of sound sources greater than the number of loudspeakers at his disposal, and to locate these sound sources precisely in 3D, even when their direction is not that of a loudspeaker. The applications of spatialized representations of sound are numerous, and include the precise localization of sound sources in 3 dimensions by a user from sound from a stereo headset, or the localization of sound sources in 3 dimensions by users in a room, the sound being produced by speakers, for example 5.1 speakers. In addition, the spatialized representations of sound allow the creation of new sound effects. For example, they allow the rotation of a sound scene or the application of reflection of a sound source to simulate the rendering of a given sound environment, for example a cinema hall or a concert hall.

Spatialized representations are carried out in two main stages: ambisonic encoding, and ambisonic decoding. To benefit from a spatialized representation of sound, real-time ambisonic decoding is always necessary. Real-time sound production or processing may also involve real-time ambisonic encoding thereof. Ambisonic encoding being a complex task, real-time ambisonic encoding capabilities may be limited. For example, a given computation capacity will only be able to encode in real time a limited number of sound sources.

Dans laquelle p ( r , t ) représente la pression sonore, à un instant t, dans la direction r par rapport au point auquel le champ sonore est calculé. j^m représente la fonction sphérique de Bessel d'ordre m.The techniques of spatialized representation of sound are described in particular by J. Daniel, Representations of acoustic fields, application to the transmission and reproduction of sound scenes in a multimedia context, INIST-CNRS, INIST call number: T 139957 . Ambisonic encoding of a sound field consists of the decomposition of the sound pressure field at a point, corresponding for example to the position of a user, in the form of spherical coordinates, expressed in the following form: $$p\left(\overrightarrow{r},t\right)={\displaystyle \sum _{m=0}^{\infty }{j}^m{j}_m\left(\mathit{kr}\right)}{\displaystyle \sum _{\mathrm{not}=-m}^{+m}{B}_{\mathit{mn}}\left(t\right)Y_{\mathit{mn}}\left(\theta ,\phi \right)}$$

In which p ( r , t ) represents the sound pressure, at a time t, in the direction r relative to the point at which the sound field is calculated. j ^m represents the spherical Bessel function of order m.

Y _mn ( θ , ϕ ) represents the spherical harmonic of order mn in the directions ( θ , ϕ ) . defined by management r . The symbol B _mn ( t ) defines the ambisonic coefficients corresponding to the various spherical harmonics, at an instant t.

The ambisonic coefficients therefore define, at each instant, the entire sound field surrounding a point. The processing of sound fields in the ambisonic domain has particularly interesting properties. In particular, it is very easy to rotate the entire sound field. It is also possible to broadcast on loudspeakers, from a set of ambisonic coefficients, sound comprising direction information. For example, it is possible to broadcast sound to 5.1-type speakers. It is also possible to play back, in a headset having only one left speaker and one right speaker, sound including direction information, using transfer functions known as HRTF ( Head-Related Transfer Functions). These functions allow a directional signal to be output to two speakers, adding to at least one channel of a stereo signal a delay and / or attenuation, which will be interpreted by the brain as defining the direction of the sound source.

The so-called HOA decomposition (from the English acronym Higher Order Ambisonics, or Higher Order Ambisonia) consists of truncating this infinite sum to an order M, greater than or equal to 1: $$p\left(\overrightarrow{r},t\right)={\displaystyle \sum _{m=0}^M{j}^m{j}_m\left(\mathit{kr}\right)}{\displaystyle \sum _{\mathrm{not}=-m}^{+m}{B}_{\mathit{mn}}\left(t\right)Y_{\mathit{mn}}\left(\theta ,\phi \right)}$$

In general, a sufficiently distant source is considered to propagate a sound wave in a spherical manner. It is then possible to consider that the value at an instant t of an ambisonic coefficient B _mn ( t ) linked to this source depends, on the one hand, on the sound pressure S ( t ) of the source at this instant t, and on the other hand of the spherical harmonic linked to the orientation ( θ _s , ϕ _s ) of this sound source. We can therefore write, for a single sound source: $$B_{\mathit{mn}}\left(t\right)=S\left(t\right)Y_{\mathit{mn}}\left({\theta }_s,{\phi }_s\right)$$

In the case of a set of N _s distant sound sources, the ambisonic coefficients describing the sound scene are calculated as the sum of the ambisonic coefficients of each of the sources, each source i having an orientation ( θ _si , ϕ _si ): $$B_{\mathit{mn}}\left(t\right)={\displaystyle \sum _{i=0}^{{\mathrm{NOT}}_s-1}{S}_i\left(t\right)Y_{\mathit{mn}}\left({\theta }_{s_i},{\phi }_{s_i}\right)}$$

Les coefficients ambisoniques gardant la forme B_mn , avec, à l'ordre M, m allant de 0 à M, et n allant de -m à m.We can also represent this calculation in the form of a vector: $$\left(\begin{array}{c}{B}_{00}\left(t\right)\\ B_{1-1}\left(t\right)\\ B_{10}\left(t\right)\\ B_{11}\left(t\right)\\ ⋮\\ B_{\mathit{MM}}\left(t\right)\end{array}\right)={\displaystyle \sum _{i=0}^{{\mathrm{NOT}}_s-1}{S}_i\left(t\right)\left(\begin{array}{c}{Y}_{00}\left({\theta }_{s_i},{\phi }_{s_i}\right)\\ Y_{1-1}\left({\theta }_{s_i},{\phi }_{s_i}\right)\\ Y_{10}\left({\theta }_{s_i},{\phi }_{s_i}\right)\\ Y_{11}\left({\theta }_{s_i},{\phi }_{s_i}\right)\\ ⋮\\ Y_{\mathit{MM}}\left({\theta }_{s_i},{\phi }_{s_i}\right)\end{array}\right)}$$

The ambisonic coefficients keeping the form B _mn , with, at order M, m ranging from 0 to M, and n ranging from -m to m.

An apparatus comprising an ambisonic encoding of at least one source can therefore define a complete sound field, by calculating the ambisonic coefficients at an order M. Depending on the order M, and the number of sources, this calculation can be long and resource-hungry. Indeed, at an order M, (M + 1) ² ambisonic coefficients are calculated at each instant t. For each coefficient, the contribution B _mn ( t ) = S ( t ) Y _mn ( θ _s , ϕ _s ) of each of the N _s sources must be calculated. If a source S is fixed, the spherical harmonic Y _mn ( θ _s , ϕ _s ) can be pre-calculated. Otherwise, it must be recalculated at each moment.

An increase of the order of the ambisonic coefficient allows a better quality of the auditory rendering. Therefore, it can be difficult to achieve good sound quality, while preserving reasonable charge, computing time, power consumption, and battery usage. This is all the more true since the calculations of ambisonic coefficients are often carried out in real time on mobile devices. This is for example the case with a smartphone, for listening to music in real time, with directional information calculated using ambisonic coefficients.

This problem is even stronger when reflections are calculated in a sound scene.

The calculation of reflections makes it possible to simulate a sound scene in a room, for example a cinema or concert hall. Under these conditions, the sound is reflected on the walls of the room, giving a characteristic "atmosphere", the reflections being defined by the respective positions of the sound sources, of the listener, but also by the materials on which the sound waves are. diffuse, for example the material of the walls. The creation of room effects using ambisonic audio encoding is described in particular by J. Daniel, Representations of acoustic fields, application to the transmission and reproduction of sound scenes in a multimedia context, INIST-CNRS, INIST code: T 139957, pp. 283-287 .

It is possible to simulate the effect of the reflections and to give an “atmosphere” in ambisonics, by adding, for each sound source, a set of secondary sound sources, the intensity and direction of which are calculated from the reflections of the sources. sound on the walls and obstacles of a soundstage. A few sound sources are necessary, for each initial sound source, in order to satisfactorily simulate a sound scene. However, this makes the aforementioned calculation and battery capacity problem even more critical, since the complexity of calculating the ambisonic coefficients is further multiplied by the number of secondary sound sources. The complexity of the calculation of the ambisonic coefficients for a satisfactory sound rendering can then make this solution impractical, for example because it becomes impossible to calculate the ambisonic coefficients in real time, because the computation load of the ambisonic coefficients becomes too important, or because the power and / or battery consumption on a mobile device becomes prohibitive.

• Evaluant la puissance de chaque source sonore ;
• Classant les sources sonores, de la plus à la moins puissante ;
• Supprimant les sources sonores les moins puissantes ;
• Groupant les sources sonores restantes par grappes de sources sonores proches les unes des autres, et les fusionnant pour obtenir, pour chaque grappe, une unique source sonore virtuelle.

N. Tsingos et al. Perceptual Audio Rendering of Complex Virtual Environment, ACM Transactions on Graphics (TOG) - Proceedings of ACM SIGGRAPH 2004, Volume 23 Issue 3, August 200, pp. 249-258 discloses a binaural processing method to overcome this problem. The solution proposed by Tsingos consists in reducing the number of sound sources by:

• Evaluating the power of each sound source;
• Classifying the sound sources, from the most to the least powerful;
• Eliminating the less powerful sound sources;
• Grouping the remaining sound sources by clusters of sound sources close to each other, and merging them to obtain, for each cluster, a single virtual sound source.

The method disclosed by Tsingos makes it possible to reduce the number of sound sources, and therefore the complexity of the overall processing when reverbs are used. However, this technique has several drawbacks. It does not improve the complexity of processing the reverbs themselves. The problem encountered would therefore arise again if, with a reduced number of sources, one wishes to increase the number of reverberations. In addition, the processing for determining the sound power of each source, and merging the sources by clusters themselves have a significant computational load. The experiments described are limited to cases where the sound sources are known in advance, and their respective powers are pre-calculated. In the case of sound scenes for which several sources of variable intensities are present, and whose powers must be recalculated, the associated calculation load would, at least partially, cancel out the calculation gain obtained by limiting the number of sources.

Finally, the tests conducted by Tsingos give satisfactory results when the sound sources can be assimilated to noise, for example in the case of a crowd in the metro. On other types of sound sources, such a method could prove to be harmful. For example, when recording a concert given by an orchestra symphonic, it is common for several instruments, although having low sound power, to contribute significantly to the overall harmony. Simply removing the associated sound sources, because they are relatively weak, would seriously damage the quality of the recording.

The document US 6021206 discloses filtering of virtual sound sources corresponding to reflections including delay and attenuation.

The document Markus Noistering et al. "A 3D AMBISONIC BASED BINAURAL SOUND REPRODUCTION SYSTEM" discloses the creation of virtual sound sources corresponding to the reflections and the application of gain and delay to each of these sources

The document US 2007/160216 generally discloses the calculation of gains as a function of the position of a sound source for binauralization.

The document US 2005/069143 discloses the application of HRTF functions to sound in the frequency domain.

The document US 2011/305344 discloses a method of transforming sound tracks before binaural encoding, in order to minimize the need for a “sweet spot”, in particular by converting certain tracks into mono.

There is therefore a need for an apparatus and a method for the calculation of the ambisonic coefficients, which makes it possible to calculate in real time a set of ambisonic coefficients representative of at least one sound source and one or more reflections thereof. in a sound scene, while limiting the additional complexity of calculation linked to the reflection (s) from the sound source, without reducing the number of sound sources a priori.

SUMMARY OF THE INVENTION

To this end, the invention relates to an ambisonic encoder for a sound wave with a plurality of reflections, comprising: a logic for frequency transformation of the sound wave; logic for calculating spherical harmonics of the sound wave and the plurality of reflections from a position of a source of the sound wave and positions of obstacles to propagation of the sound wave; a plurality of filtering logics in the frequency domain receiving as input spherical harmonics of the plurality of reflections, each filtering logic consisting of an attenuation and a delay of a reflection, and being parameterized by an acoustic coefficient and a delay of said reflection; a logic of adding spherical harmonics of the sound wave and of the outputs of the filtering logics, into a set of spherical harmonics representative both of the sound wave and of the plurality of reflections in the frequency domain; a logic for multiplying said set of spherical harmonics representative both of the sound wave and of the plurality of reflections in the frequency domain by sound intensity values of the wave at the output of the frequency transformation, in order to obtaining a set of ambisonic coefficients representative of both the sound wave and the plurality of reflections.

Advantageously, the logic for calculating spherical harmonics of the sound wave is configured to calculate the spherical harmonics of the sound wave and of the plurality of reflections from a fixed position of the source of the sound wave.

Advantageously, the logic for calculating spherical harmonics of the sound wave is configured to iteratively calculate the spherical harmonics of the sound wave and of the plurality of reflections from successive positions of the source of the sound wave.

Advantageously, each reflection is characterized by a single acoustic coefficient.

Advantageously, each reflection is characterized by an acoustic coefficient for each frequency of said frequency sampling.

Advantageously, the reflections are represented by virtual sound sources.

Advantageously, the ambisonic encoder further comprises a logic for calculating the acoustic coefficients, the delays and the position of the virtual sound sources of the reflections, said calculation logic being configured to calculate the acoustic coefficients and the delays of the reflections as a function of '' estimates of a difference in the distance traveled by sound between the position of the source of the sound wave and an estimated position of a user on the one hand, and of a distance traveled by the sound between the positions of the virtual sound sources of the reflections and the estimated position of the user on the other hand.

Advantageously, the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is further configured to calculate the acoustic coefficients of the reflections as a function of at least one acoustic coefficient of at least one obstacle to propagation. of sound waves, on which the sound is reflected.

Advantageously, the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is further configured to calculate the acoustic coefficients of the reflections as a function of an acoustic coefficient of at least one obstacle to the propagation of sound waves, on which sound is reflected.

Advantageously, the logic for calculating spherical harmonics of the sound wave and of the plurality of reflections is further configured to calculate spherical harmonics of the sound wave and of the plurality of reflections at each output frequency of the transformation circuit. frequency, said ambisonic encoder further comprising a logic for calculating binaural coefficients of the sound wave, configured to calculate binaural coefficients of the sound wave by multiplying at each output frequency of the frequency transformation circuit of the sound wave the signal of the sound wave by the spherical harmonics of the sound wave and the plurality of reflections at this frequency.

Advantageously, the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is configured to calculate the acoustic coefficients and the delays of a plurality of late reflections.

The invention also relates to a method of ambisonically encoding a sound wave with a plurality of reflections, as defined by claim 12.

The invention also relates to a computer program for ambisonic encoding of a plurality of reflections sound wave, as defined by claim 13.

The ambisonic encoder according to the invention makes it possible to improve the feeling of immersion in a 3D audio scene.

The complexity of encoding the reflections of sound sources from an ambisonic encoder according to the invention is less than the complexity of encoding the reflections of sound sources from an ambisonic encoder according to the state of the art.

The ambisonic encoder according to the invention makes it possible to encode a greater number of reflections from a sound source in real time.

The ambisonic encoder according to the invention makes it possible to reduce the power consumption associated with ambisonic encoding, and to increase the life of a battery of a mobile device used for this application.

LIST OF FIGURES

• les figures 1a et 1b, deux exemples de systèmes d'écoute d'onde sonore, selon deux modes de réalisation de l'invention
• la figure 2, un exemple d'un système de binauralisation comprenant un moteur de binauralisation par source sonore d'une scène audio selon l'état de l'art ;
• les figures 3a et 3b, deux exemples de moteurs de binauralisation d'une scène 3D, respectivement dans le domaine temporel et le domaine fréquentiel selon l'état de l'art
• la figure 4, un exemple d'encodeur ambisonique d'une onde sonore à une pluralité de réflexions, dans un ensemble de modes de mise en œuvre de l'invention ;
• la figure 5, un exemple de calcul d'une source sonore secondaire, dans un mode de mise en œuvre de l'invention ;
• la figure 6, un exemple de calcul de réflexions précoces et de réflexions tardives, dans un mode de réalisation de l'invention ;
• la figure 7, une méthode d'encodage d'une onde sonore à une pluralité de réflexions dans un ensemble de modes de mise en œuvre de l'invention.

Other characteristics will appear on reading the detailed description given by way of example and not limiting which follows, given with reference to the appended drawings which represent:

• the figures 1a and 1b , two examples of sound wave listening systems, according to two embodiments of the invention
• the figure 2 , an example of a binauralization system comprising a binauralization engine by sound source of an audio scene according to the state of the art;
• the figures 3a and 3b , two examples of binauralization engines of a 3D scene, respectively in the time domain and the frequency domain according to the state of the art
• the figure 4 , an example of an ambisonic encoder of a sound wave with a plurality of reflections, in a set of embodiments of the invention;
• the figure 5 , an example of calculation of a secondary sound source, in one embodiment of the invention;
• the figure 6 , an example of calculation of early reflections and late reflections, in one embodiment of the invention;
• the figure 7 , a method of encoding a sound wave with a plurality of reflections in a set of embodiments of the invention.

DETAILED DESCRIPTION

The figures 1a and 1b show two examples of sound wave listening systems, according to two embodiments of the invention.

The figure 1a represents an example of a sound wave listening system, according to one embodiment of the invention.

The system 100a comprises a touch pad 110a, a headset 120a to allow a user 130a to listen to a sound wave. The system 100a comprises, by way of example only, a touchscreen tablet. However, this example is also applicable to a smartphone, or to any other mobile device having display and sound broadcasting capabilities. The sound wave can for example come from playing a movie or a game. According to several embodiments of the invention, the system 100a can be configured to listen to several sound waves. For example, when the system 100a is configured for playing a movie comprising a 5.1 multichannel sound track, 6 sound waves are listened to simultaneously. Likewise, when system 100a is configured to play a game, many sound waves can be heard simultaneously. For example, in the case of a game involving several characters, a sound wave can be created for each character.

Each of the sound waves is associated with a sound source, the position of which is known.

The touchscreen tablet 110a comprises an ambisonic encoder 111a according to the invention, a transformation circuit 112a, and an ambisonic decoder 113a.

According to a set of embodiments of the invention, the ambisonic encoder 111a, the transformation circuit 112a and the ambisonic decoder 113a consist of computer code instructions executed on a processor of the touch pad. They may for example have been obtained by installing a specific application or software on the tablet. In other embodiments of the invention, at least one of the ambisonic encoder 111a, the transformation circuit 112a and the ambisonic decoder 113a is a specialized integrated circuit, for example an ASIC (acronym for English “Application-Specific Integrated Circuit, literally“ application-specific integrated circuit ”), an FPGA (acronym for English Field-Programmable Gate Array).

The ambisonic encoder 111a is configured to calculate, in the frequency domain, a set of ambisonic coefficients representative of the whole of a sound scene, from at least one sound wave. It is further configured to apply reflections to at least one sound wave, in order to simulate a listening environment, for example a movie theater of a certain size, or a concert hall.

The transformation circuit 112a is configured to perform rotations of the soundstage by modifying the ambisonic coefficients, in order to simulate the rotation of the user's head, so that, whatever the orientation of his face, the different sound waves seem to come from the same position. For example, if the user turns his head to the left by an angle α, a rotation of the sound stage to the right by the same angle α makes it possible to continue sending the sound to him always from the same direction. According to a set of embodiments of the invention, the helmet 120a is equipped with at least one movement sensor 121a, for example a gyrometer, making it possible to obtain an angle or a derivative of an angle of rotation of the head. user 130a. A signal representative of an angle of rotation, or of a derivative of an angle of rotation, is then sent by the headphones 121a to the tablet 120a, so that the transformation circuit 112a performs the rotation of the corresponding sound scene.

The ambisonic decoder 113a is configured to reproduce the sound scene on the two stereo channels of the headphones 120a, by converting the transformed ambisonic coefficients into two stereo signals, one for the left channel and the other for the right channel. In a set of embodiments of the invention, the ambisonic decoding is carried out using functions called HRTF (acronym for the English “Head Related Transfer Functions”, literally Head Related Transfer Functions) making it possible to reproduce the directions of the different sound sources on two stereo channels. French patent application no. 1558279 , filed by the applicant, describes a method for creating HRTF functions optimized for a user as a function of a bank of HRTF functions, and of the facial characteristics of said user.

The system 100a thus allows its user to benefit from a particularly immersive experience: during a game or a multimedia content reading, in addition to the image, this system allows him to benefit from an impression. immersion in a sound scene. This impression is amplified both by following the orientations of the different sound sources when the user turns his head, and by the application of reflections giving an impression of immersion in a particular listening environment. This system makes it possible, for example, to watch a film or a concert with an audio headset, while having an impression of immersion in a cinema hall or a concert hall. All of these operations are carried out in real time, which makes it possible to constantly adapt the sound perceived by the user to the orientation of his head.

The ambisonic encoder 111a according to the invention makes it possible to encode a greater number of reflections from sound sources, with less complexity compared to an ambisonic encoder of the prior art. It therefore makes it possible to perform all the ambisonic calculations in real time, while increasing the number of reflections from sound sources. This increase in the number of reflections makes it possible to model more precisely the simulated listening environment (concert hall, cinema, etc.) and therefore improve the feeling of immersion in the sound scene. The reduction in the complexity of the ambisonic encoding also makes it possible, by considering an identical number of sound sources, to reduce the electrical consumption of the encoder compared to an encoder of the state of the art, and therefore to increase the battery discharge time of the touch pad 110a. This therefore allows the user to enjoy multimedia content for a longer period of time.

The figure 1b represents a second example of a sound wave listening system, according to one embodiment of the invention.

The system 100b includes a central unit 110b connected to a screen 114b, a mouse 115b and a keyboard 116b and a headset 120b and is used by a user 130b. The central unit comprises an ambisonic encoder 111b according to the invention, a transformation circuit 112b, and an ambisonic decoder 113b, respectively similar to the ambisonic encoder 111a, transformation circuit 112a, and ambisonic decoder 113a of the system 100a. Similarly to the system 100a, the ambisonic encoder 111b is configured to encode at least one wave representative of a sound scene by adding reflections thereto, the headphones 120a include at least one movement sensor 120b, the transformation circuit 120b is configured to perform soundstage rotations to follow the orientation of the user's head, and the ambisonic decoder 113b is configured to output sound on the two stereo channels of the headphones 120b, so that the user 130b has an impression of immersion in a sound scene.

The 100b system is suitable for viewing multimedia content, but also for video games. Indeed, in a video game, very many sound waves, coming from different sources, can occur. This is for example the case in a strategy or war game, in which many characters can emit different sounds (noises of footsteps, running, shots, etc.) for various sound sources. A 111b ambisonic encoder can encode all these sources, while adding many reflections to them making the scene more realistic and immersive, in real time. Thus, the system 100b comprising an ambisonic encoder 111b according to the invention allows an immersive experience in a video game, with a large number of sound sources and reflections.

The figure 2 represents an example of a binauralization system comprising a binauralization engine by sound source of an audio scene according to the state of the art.

The binauralization system 200 is configured to transform a set 210 of sound sources of a soundstage into a left channel 240 and a right channel 241 of a stereo listening system, and includes a set of binaural motors 220, including a binaural motor by sound source.

The sources can be from any type of sound source (mono, stereo, 5.1, multiple sound sources in the case of a video game for example). Each sound source is associated with an orientation in space, for example defined by angles ( θ , ϕ ) in a frame of reference, and by a sound wave, itself represented by a set of temporal samples.

• effectuer un encodage HOA de la source sonore à un ordre M ;
• effectuer une transformation sur les coefficients binauraux, par exemple une rotation ;
• calculer une intensité sonore p( r ,t) à des instants t pour un ensemble de canaux de sortie, dans laquelle r représente l'orientation du canal de sortie.

Each of the binauralization engines of the set 220 is configured for, for a sound source and at each instant t corresponding to a sample of the sound source:

• performing HOA encoding of the sound source at an M order;
• performing a transformation on the binaural coefficients, for example a rotation;
• calculate a sound intensity p ( r , t ) at times t for a set of output channels, in which r represents the orientation of the output channel.

The possible output channels correspond to the different listening channels, for example we can have two output channels in a stereo listening system, 6 output channels in a 5.1 listening system, etc ...

Each binauralization motor produces two outputs (one left and one right output), and the system 200 includes an addition circuit 230 of all the left outputs and an addition circuit 231 of all the right outputs of the set 220. binauralization engines. The outputs of the addition logic 230 and 231 are respectively the sound wave of the left channel 240 and the sound wave of the right channel 241 of a stereo listening system.

The system 200 makes it possible to transform the set of sound sources 210 into two stereo channels, while being able to apply all the transformations allowed by the ambisonia, such as rotations.

However, the system 200 has a major drawback in terms of calculation time: it requires calculations to calculate the ambisonic coefficients of each sound source, calculations for the transformations of each sound source, and calculations for the outputs associated with each source. sound. The computational load for the processing of a sound source by the system 200 is therefore proportional to the number of sound sources, and can, for a large number of sound sources, become prohibitive.

The figures 3a and 3b represent two examples of engines for binauralization of a 3D scene, respectively in the time domain and the frequency domain according to the state of the art.

The figure 3a represents an example of a binauralization engine of a 3D scene, in the time domain according to the state of the art.

In order to limit the complexity of the binaural processing in the case of a large number of sources, the binauralization engine 300a comprises a single HOA encoding engine 320a for all of the sources 310 of the sound scene. This encoding engine 320a is configured to calculate, at each time step, the binaural coefficients of each sound source as a function of the intensity and the position of the sound source at said time step, then to sum the binaural coefficients of the different sound sources. This makes it possible to obtain a single set 321a of binaural coefficients representative of the whole of the sound scene.

The binauralization engine 320a then comprises a coefficient transformation circuit 330a, configured to transform the set of coefficients 321a representative of the sound scene into a set of transformed coefficients 331a representative of the whole of the sound scene. This makes it possible for example to perform a rotation of the whole of the sound scene.

The binauralization engine 300a finally comprises a binaural decoder 340a, configured to restore the transformed coefficients 331a into a set of output channels, for example a left channel 341a and a right channel 342a of a stereo system.

The binauralization engine 300a therefore makes it possible to reduce the computational complexity necessary for the binaural processing of a sound scene compared to the system 200, by applying the transformation and decoding steps to the whole of the sound scene, rather than to each sound source taken individually.

figure 3b represents an example of a binauralization engine of a 3D scene, in the frequency domain according to the state of the art.

The 300b binauralization engine is quite similar to the 300a binauralization engine. It comprises a set 311b of frequency transformation logics, the set 311b comprising a frequency transformation logic for each sound source. The frequency transformation logics can for example be configured to apply a fast Fourier transform (or FFT, from the acronym Fast Fourier Transform), in order to obtain a set 312b of sources in the frequency domain. The application of frequency transforms is well known to those skilled in the art, and is for example described by A. Mertins, Signal Analysis: Wavelets, Filter banks, Time-Frequency Transforms and Applications, English (revised edition). ISBN: 9780470841839 . It consists for example in transforming, by time windows, the sound samples into frequency intensity, according to frequency sampling. The inverse operation, or inverse frequency transformation (known as FFT ^-1 or inverse fast Fourier transformation in the case of a fast Fourier transform) makes it possible to restore, from a sampling of frequencies, the intensities of sound samples .

The binauralization engine 300b then includes an HOA encoder 320b in the frequency domain. The encoder 320b is configured to calculate, for each source and at each frequency of the frequency sampling, the corresponding ambisonic coefficients, then to add the ambisonic coefficients of the different sources, in order to obtain a set 321b of ambisonic samples representative of the whole soundstage, at different frequencies. An ambisonic coefficient at a frequency f of the frequency sampling is obtained, so similar to an ambisonic coefficient at time t, by the formula: B _mn ( f ) = S ( f ) Y _mn ( θ _s , ϕ _s ).

The binauralization engine 300b then comprises a transformation circuit 330b, similar to the transformation circuit 330a, making it possible to obtain a set 331b of transformed ambisonic coefficients representative of the whole of the sound scene, and a binaural decoder 340b, configured to restore two stereo channels 341b and 342b. The binaural decoder 340b comprises an inverse frequency transformation circuit, in order to restore the stereo channels in the time domain.

The properties of the binauralization engine 300b are quite similar to those of the binauralization engine 300a. It also makes it possible to perform binaural processing of a sound scene, with reduced complexity compared to the system 200.

In the event of a significant increase in the number of sources, the complexity of the binaural processing of the binauralization engines 300a and 300b is mainly due to the calculation of the HOA coefficients by the encoders 320a and 320b. Indeed, the number of coefficients to be calculated is proportional to the number of sources. On the contrary, the transformation circuits 330a and 330b, as well as the binaural decoders 340a and 340b process sets of binaural coefficients representative of the whole of the sound scene, the number of which does not vary according to the number of sources.

For the processing of reflections, the complexity of binaural encoders 320a and 320b can increase significantly. Indeed, the state of the art solution for processing reflections consists of adding a virtual sound source for each reflection. The complexity of the HOA encoding of these encoders according to the state of the art therefore increases proportionally as a function of the number of reflections per source, and can become problematic when the number of reflections becomes too large.

The figure 4 represents an example of an ambisonic encoder of a sound wave with a plurality of reflections, in a set of embodiments of the invention.

The ambisonic encoder 400 is configured to encode a sound wave 410 with a plurality of reflections, into a set of one-order ambisonic coefficients. To do this, the ambisonic encoder is configured to calculate a set 460 of representative spherical harmonics. the sound wave and the plurality of reflections. The ambisonic encoder 400 will be described, by way of example, for the encoding of a single sound wave. However, an ambisonic encoder 400 according to the invention can also encode a plurality of sound waves, the elements of the ambisonic encoder being used in the same way for each additional sound wave. The sound wave 410 can correspond for example to a channel of an audio track, or to a dynamically created sound wave, for example a sound wave corresponding to an object of a video game. In one set of embodiments of the invention, the sound waves are defined by successive sound intensity samples. According to different embodiments of the invention, the sound waves can for example be sampled at a frequency of 22500Hz, 12000Hz, 44100 Hz, 48000 Hz, 88200 Hz, or 96000 Hz, and each of the intensity samples coded to 8, 12, 16, 24 or 32 bits. In the event of a plurality of sound waves, these can be sampled at different frequencies, and the samples can be encoded on different numbers of bits.

The ambisonic encoder 400 includes logic 420 for frequency transformation of the sound wave. This is similar to the logic 311b of frequency transformation of the sound waves of the binauralization system 300b according to the state of the art. In embodiments with a plurality of sound waves, the encoder 400 includes frequency transformation logic for each sound wave. At the output of the frequency transformation logic, a sound wave is defined 421, for a time window, by a set of intensities at different frequencies of a frequency sampling. In a set of embodiments of the invention, the frequency transformation logic 420 is an application logic of an FFT.

The encoder 400a also includes logic 430 for calculating spherical harmonics of the sound wave and the plurality of reflections from a position of a source of the sound wave and from positions of obstacles to propagation. of the sound wave. In a set of modes embodiment of the invention, the position of the source of the sound wave is defined by angles ( θ _{s i} , ϕ _{s i} ) and a distance from a listening position of the user. The calculation of spherical harmonics Y ₀₀ ( θ _{s i} , ϕ _{s i} ), Y _1-1 ( θ _{s i} , ϕ _{s i} ) Y ₁₀ ( θ _{s i} , ϕ _{s i} ), Y ₁₁ ( θ _{s i} , ϕ _{s i} ), ..., Y _MM ( θ _{s i} , ϕ _{s i} ), the sound wave at the order M can be carried out according to the methods known in the state of the art, from the angles ( θ _{s i} , ϕ _{s i} ) defining the orientation of the source of the sound wave.

Logic 430 is also configured to calculate, from the position of the source of the sound wave, a set of spherical harmonics of the plurality of reflections. In one set of embodiments of the invention, logic 430 is configured to calculate, from the position of the source of the sound wave, and positions of obstacles to the propagation of the sound wave, a orientation of a virtual source of a reflection, defined by angles ( θ _{s, r} , ϕ _{s, r} ) then, from these angles, spherical harmonics Y ₀₀ ( θ _{s, r} , ϕ _{s, r} ) , Y _1-1 ( θ _{s, r} , ϕ _{s, r} ), Y ₁₀ ( θ _{s, r} , ϕ _{s, r} ), Y ₁₁ ( θ _{s, r} , ϕ _{s, r} ), ..., Y _MM ( θ _{s, r} , ϕ _{s, r} ) of the sound wave reflection. This makes it possible to obtain, for each reflection, the spherical harmonics corresponding to the direction of the wave reflected on the obstacles to the propagation of the sound wave.

The ambisonic encoder 400 also includes a plurality 440 of filter logic in the frequency domain receiving as input spherical harmonics of the plurality of reflections, each filter logic being parameterized by acoustic and delay coefficients of the reflections. In the remainder of the description, α _r will be called an acoustic coefficient of a reflection and δ _r a delay of a reflection. According to various embodiments of the invention, the acoustic coefficient can be a coefficient of α _r reverberation, representative of a ratio of the intensities of a reflection to the intensities of the sound source and defined between 0 and 1. According to d ' other embodiments of the invention, the acoustic coefficient is a coefficient α _a said of attenuation or absorption, that is to say a coefficient defined between 0 and 1 such that α _a = α _r - 1. These filtering logics allow d 'apply a delay and an attenuation to the ambisonic coefficients of a reflection. Thus, the combination of the orientation of the virtual source of the reflection, the delay and the attenuation of the reflection makes it possible to model each reflection as a replica of the sound source, coming from a different direction, assigned a delay and attenuated, following the path and reflections of the sound wave. This modeling allows, with several reflections, to simulate the propagation of a sound wave in a scene in a simple and efficient way.

In general, the filtering, at a frequency f, of a spherical harmonic of a reflection can be written: H _r (f) Y _ij ( θ _{s, r} , ϕ _{s, r} ). In one embodiment of the invention, a filter logic 440 is configured to filter the spherical harmonics by applying: α _r e ^{- j 2 πδ r} Y _ij ( θ _{s, r} , ϕ _{s, r} ). In this embodiment, the coefficient α _r is treated as a reverberation coefficient. In other embodiments, a coefficient α _a can be treated as an attenuation coefficient, and the filtering of spherical harmonics can for example be performed by applying: ( 1 - α _a ) e ^{-j 2 πfδ r} Y _ij ( θ _{s, r} , ϕ _{s, r} ). In the remainder of the description, unless otherwise specified, the coefficient α _{r will be considered} as a reverberation coefficient. A person skilled in the art could however easily implement the various embodiments of the invention with an attenuation coefficient rather than a reverberation coefficient.

${\mathbf{Y}}_{\mathit{ij}}\left({\theta }_{{\mathbf{s}}_{\mathbf{i}}},{\phi }_{{\mathbf{s}}_{\mathbf{i}}}\right)+{\displaystyle {\sum }_{\mathbf{r}=0}^{{\mathbf{N}}_{\mathbf{r}}}{\mathbf{H}}_{\mathbf{r}}\left(\mathbf{f}\right){\mathbf{Y}}_{\mathit{ij}}\left({\theta }_{\mathbf{s},\mathbf{r}},{\phi }_{\mathbf{s},\mathbf{r}}\right),}$

dans laquelle Y_ij ( θ_si , ϕ_si ) est une harmonique sphérique de la source de l'onde sonore, N_r est le nombre de réflexions de l'onde sonore, Y_ij ( θ_s,r , ϕ_s,r ) sont les harmoniques sphériques des positions des sources sonores virtuelles des réflexions, et les termes H_r(f) sont les logiques de filtrage des harmoniques sphériques pour la réflexion r à une fréquence f. Dans un ensemble de modes de réalisation de l'invention, les logiques de filtrage H_r(f) sont telles que H_r(f) = α_re ^-j2πfδ_r , et les harmoniques sphériques Y_ij à l'ordre M, représentatives à la fois de l'onde sonore, et des réflexions de l'onde sonore sont égales, en sortie de la logique 450 d'addition, à: $\mathit{Y}{\prime }_{\mathit{ij}}={\mathit{Y}}_{\mathit{ij}}\left({\theta }_{{\mathit{s}}_{\mathit{i}}},{\phi }_{{\mathit{s}}_{\mathit{i}}}\right)+{\displaystyle {\sum }_{\mathit{r}=0}^{{\mathit{N}}_{\mathit{r}}}{\alpha }_{\mathit{r}}{\mathit{e}}^{-\mathit{j}2\pi \mathit{f}{\delta }_{\mathit{r}}}{\mathit{Y}}_{\mathit{ij}}\left({\theta }_{\mathit{s},\mathit{r}},{\phi }_{\mathit{s},\mathit{r}}\right)}.$

The ambisonic encoder 400 also includes logic 450 for adding the spherical harmonics of the sound wave and the outputs of the filtering logic. This logic makes it possible to obtain a set Y _'00 , Y' _1-1 , Y _'10 , Y' ₁₁ , ... Y ' _MM of spherical harmonics at order M, representative of both the wave sound, and reflections of the sound wave, in the frequency domain. A spherical harmonic Y ' _ij (with 0 ≤ i ≤ M, and -i ≤ j ≤ i) representative both of the sound wave, and of the reflections of the sound wave, is therefore equal, at the output of the logic addition 450, to the value ${\mathbf{Y}}_{\mathit{ij}}=$

${\mathbf{Y}}_{\mathit{ij}}\left({\theta }_{{\mathbf{s}}_{\mathbf{i}}},{\phi }_{{\mathbf{s}}_{\mathbf{i}}}\right)+{\displaystyle {\sum }_{\mathbf{r}=0}^{{\mathbf{NOT}}_{\mathbf{r}}}{\mathbf{H}}_{\mathbf{r}}\left(\mathbf{f}\right){\mathbf{Y}}_{\mathit{ij}}\left({\theta }_{\mathbf{s},\mathbf{r}},{\phi }_{\mathbf{s},\mathbf{r}}\right),}$

where Y _ij ( θ _si , ϕ _si ) is a spherical harmonic of the source of the sound wave, N _r is the number of reflections of the sound wave, Y _ij ( θ _{s, r} , ϕ _{s, r} ) are the spherical harmonics of the positions of the virtual sound sources of the reflections, and the terms H _r (f) are the filtering logic of the spherical harmonics for the reflection r at a frequency f. In a set of embodiments of the invention, the filtering logics H _r (f) are such that H _r (f) = α _r e ^{-j 2 πfδ _r} , and the spherical harmonics Y _ij to the order M , representative of both the sound wave, and the reflections of the sound wave are equal, at the output of addition logic 450, to: $\mathit{Y}{\prime }_{\mathit{ij}}={\mathit{Y}}_{\mathit{ij}}\left({\theta }_{{\mathit{s}}_{\mathit{i}}},{\phi }_{{\mathit{s}}_{\mathit{i}}}\right)+{\displaystyle {\sum }_{\mathit{r}=0}^{{\mathit{NOT}}_{\mathit{r}}}{\alpha }_{\mathit{r}}{\mathit{e}}^{-\mathit{j}2\pi \mathit{f}{\delta }_{\mathit{r}}}{\mathit{Y}}_{\mathit{ij}}\left({\theta }_{\mathit{s},\mathit{r}},{\phi }_{\mathit{s},\mathit{r}}\right)}.$

According to different embodiments of the invention, the number N _r of reflections can be predefined. According to other embodiments of the invention, the reflections of the sound wave are preserved according to their acoustic coefficient, the number Nr of reflections then depending on the position of the sound source, on the position of the user, and obstacles to the propagation of sound. In the example above, the acoustic coefficient is defined as a ratio of the intensity of the reflection to the intensity of the sound source, or a reverberation coefficient. In one embodiment of the invention, the reflections of the sound wave having an acoustic coefficient greater than or equal to a predefined threshold are preserved. In other embodiments, the acoustic coefficient is defined as an attenuation coefficient, i.e. a ratio between the sound intensity absorbed by the obstacles to the propagation of sound waves and the air path and intensity. sound source. In this embodiment, the reflections of the sound wave having an acoustic coefficient less than or equal to a predefined threshold are preserved

Thus, the ambisonic encoder 400 makes it possible to calculate a set of spherical harmonics Y ' _ij representative both of the sound wave and of its reflections. Once these spherical harmonics have been calculated, the encoder can include a logic of multiplying the spherical harmonics by the sound intensity values of the source at the different frequencies, in order to obtain ambisonic coefficients representative of both the sound wave and reflections. In embodiments with several sound sources, the encoder 400 includes logic for adding the ambisonic coefficients of the different sound sources and their reflections, making it possible to obtain at the output ambisonic coefficients representative of the entire sound scene. .

In a set of embodiments of the invention, the M-order ambisonic coefficients representative of the sound scene are then equal, at the output of the logic for adding the ambisonic coefficients of the different sound sources and their reflections, for Ns sound sources and for a frequency f, at: $$\left(\begin{array}{c}{B}_{00}\left(f\right)\\ B_{1-1}\left(f\right)\\ B_{10}\left(f\right)\\ B_{11}\left(f\right)\\ ⋮\\ B_{\mathit{MM}}\left(f\right)\end{array}\right)={\displaystyle \sum _{i=0}^{{\mathrm{NOT}}_s-1}{S}_i\left(f\right)\left(\begin{array}{c}{Y}_{00}\left({\theta }_{s_i},{\phi }_{s_i}\right)+{\displaystyle \sum _{r=0}^{{\mathrm{NOT}}_r}{\mathit{H}}_{\mathit{r}}\left(\mathit{f}\right)Y_{00}\left({\theta }_{s,r},{\phi }_{s,r}\right)}\\ Y_{1-1}\left({\theta }_{s_i},{\phi }_{s_i}\right)+{\displaystyle \sum _{r=0}^{{\mathrm{NOT}}_r}{\mathit{H}}_{\mathit{r}}\left(\mathit{f}\right)Y_{1-1}\left({\theta }_{s,r},{\phi }_{s,r}\right)}\\ Y_{10}\left({\theta }_{s_i},{\phi }_{s_i}\right)+{\displaystyle \sum _{r=0}^{{\mathrm{NOT}}_r}{\mathit{H}}_{\mathit{r}}\left(\mathit{f}\right)Y_{10}\left({\theta }_{s,r},{\phi }_{s,r}\right)}\\ Y_{11}\left({\theta }_{s_i},{\phi }_{s_i}\right)+{\displaystyle \sum _{r=0}^{{\mathrm{NOT}}_r}{\mathit{H}}_{\mathit{r}}\left(\mathit{f}\right)Y_{11}\left({\theta }_{s,r},{\phi }_{s,r}\right)}\\ ⋮\\ Y_{\mathit{MM}}\left({\theta }_{s_i},{\phi }_{s_i}\right)+{\displaystyle \sum _{r=0}^{{\mathrm{NOT}}_r}{\mathit{H}}_{\mathit{r}}\left(\mathit{f}\right)Y_{\mathit{MM}}\left({\theta }_{s,r},{\phi }_{s,r}\right)}\end{array}\right)}$$

The use of a single ambisonic coefficient Y ' _ij representative of both the sound wave and its reflections makes it possible to significantly reduce the calculation operations making it possible to obtain the ambisonic coefficients, especially when the number of reflections is Student. Indeed, this makes it possible to reduce the number of multiplications, since it is no longer necessary to multiply each of the intensities S _i ( f ) of a source for each frequency by each of the spherical harmonics Y _ij ( θ _{s, r} , ϕ _{s, r} ), for each value of i such that 0 ≤ i ≤ M, each value of j such that -i ≤ j ≤ i, and each reflection. This reduction in the number of multiplications allows a significant reduction in the complexity of the computation, particularly in the case of a high number of reflections.

In one set of embodiments of the invention, the sound wave spherical harmonics calculation logic 430 is configured to calculate the spherical harmonics of the sound wave and the plurality of reflections from a position fixed source of the sound wave. In this case, the orientations ( θ _si , ϕ _si ) of the sound source, and the orientations ( θ _{s, r} , ϕ _{s, r} ) of each of the harmonics are constant. The spherical harmonics of the sound wave and of the plurality of reflections then also have a constant value, and can be calculated only once for the sound wave.

In other embodiments of the invention, the sound wave spherical harmonics calculation logic 430 is configured to iteratively calculate the spherical harmonics of the sound wave and the plurality of reflections from successive positions of the source of the sound wave. According to different embodiments of the invention, different possibilities exist for defining the iterations of calculation. In one embodiment of the invention, logic 430 is configured to recalculate the values of the spherical harmonics of the sound wave and of the plurality of reflections each time a change in the position of the source of the wave. sound or user's position is detected. In another embodiment of the invention, logic 430 is configured to recalculate the values of the spherical harmonics of the sound wave and of the plurality of reflections at regular intervals, for example every 10 ms. In another embodiment of the invention, logic 430 is configured to recalculate the values of the spherical harmonics of the sound wave and of the plurality of reflections at each of the time windows used by the frequency transformation logic 420 of the sound wave. sound wave to convert the temporal samples of the sound wave into frequency samples.

In a set of embodiments of the invention, each reflection is characterized by a single acoustic coefficient α _r .

In other embodiments of the invention, each reflection is characterized by an acoustic coefficient for each frequency of said frequency sampling. This makes it possible to obtain different acoustic coefficients for the different frequencies, and to improve the rendering of certain effects. For example, it is known that thick materials absorb low frequencies more significantly. Likewise, certain types of materials absorb and reflect high frequencies in different ways. Thus, the definition of different acoustic coefficients for the same reflection and different frequencies makes it possible to characterize the materials encountered by the reflections, allowing a better rendering of different types of room, according to the materials of the walls of this one.

In a set of embodiments of the invention, a reflection at a frequency can be considered as zero, as a function of a comparison between the acoustic coefficient α _r for this frequency and a predefined threshold. For example, if the coefficient α _r represents a reverberation coefficient, the frequency is considered to be zero if it is less than a predefined threshold. On the contrary, if it is an attenuation coefficient, the frequency is considered zero if it is greater than or equal to a predefined threshold. This makes it possible to further limit the number of multiplications, and therefore the complexity of the ambisonic encoding, while having a minimal impact on the binaural rendering.

In a set of embodiments of the invention, the ambisonic encoder 400 includes logic for calculating the acoustic coefficients and the delays, and the position of the virtual sound source of the reflections. This calculation logic can for example be configured to calculate the acoustic coefficients and the delays of the reflections as a function of estimates of a difference in distance traveled by the sound between the position of the source of the sound wave and an estimated position d 'a user on the one hand, and the distance traveled by the sound between the positions of the virtual sound sources of the reflections and the estimated position of the user on the other hand. It is indeed easy, knowing the difference in distance traveled by the sound wave to reach the user, in a straight line from the sound source on the one hand, and through a reflection on the other hand, and knowing the speed of the sound, to deduce the delay felt by the user between the sound coming from the sound source in a straight line on the one hand, and the sound having been affected by the reflection on the other hand.

Likewise, it is known that the intensity of a sound wave decreases as it travels through the air. The logic for calculating the acoustic coefficients and the delays, and the position of the virtual sound source of the reflections can therefore be configured to calculate an acoustic coefficient of a reflection of the sound wave as a function of the difference in distance traveled between the sound from the sound source in a straight line on the one hand, and sound having been affected by reflection on the other hand.

In other embodiments of the invention, the logic for calculating the acoustic coefficients and delays, and the position of the virtual sound source of the reflections is also configured to calculate the acoustic coefficients of the reflections as a function of a coefficient acoustics of at least one obstacle to the propagation of sound waves, on which the sound is reflected. This makes it possible to better model the absorption by the materials of a room, and the acoustic coefficient of the obstacle can be variable according to the different frequencies. The acoustic coefficient of the obstacle can be a reverberation coefficient or an attenuation coefficient.

The figure 5 represents an example of calculation of a secondary sound source, in one embodiment of the invention.

In this example a source of the sound wave has a position 520 in a room 510, and the user has a position 540. The room 510 consists of 4 walls 511, 512, 513 and 514.

In a set of embodiments of the invention, the logic for calculating the acoustic coefficients and the delays, and the position of the virtual sound source of the reflections is configured to calculate the position, delay and attenuation of the virtual sound sources. reflections as follows: for each of the walls 511, 512, 513, 514, the logic is configured to calculate a position of a virtual sound source of a reflection as the symmetrical of the position of the sound source with respect to a wall. The calculation logic is thus configured to calculate the positions 521, 522, 523 and 524 of four virtual sound sources of the reflections, respectively with respect to the walls 511, 512, 513 and 514.

For each of these virtual sound sources, the calculation logic is configured to calculate a path of travel of the sound wave, and to deduce therefrom the corresponding acoustic coefficient and the corresponding delay. For example, in the case of the virtual sound source 511, the sound wave follows the path 530 to point 531 of the wall 512, then the path 532 to the position of the user 540. The distance traveled by the user. sound along the path 530, 532 makes it possible to calculate an acoustic coefficient and a delay of the reflection. In a set of embodiments of the invention the calculation logic is also configured to apply an acoustic coefficient corresponding to the absorption of the wall 512 at point 531. In a set of embodiments of the invention, this coefficient depends different frequencies, and can for example be determined, for each frequency, as a function of the material and / or the thickness of the wall 512.

In one set of embodiments of the invention, the virtual sound sources 521, 522, 523, 524 are used to calculate secondary virtual sound sources, corresponding to multiple reflections. For example, a secondary virtual source 533 can be calculated as the symmetrical of the virtual source 521 with respect to the wall 514. The corresponding sound wave path then comprises the segments 530 up to point 531; 534 between points 531 and 535; 536 between point 535 and position 540 of the user. The acoustic coefficients and the delays can then be calculated from the distance traveled by the sound on segments 531, 535 and 536, and the absorption of the walls at points 531 and 535.

According to different embodiments of the invention, virtual sound sources corresponding to reflections can be calculated up to a predefined order n. Different embodiments are possible to determine which reflections to keep. In one embodiment of the invention, the calculation logic is configured to calculate, for each virtual sound source, a higher order virtual sound source for each of the walls, up to a predefined order n. In one embodiment, the ambisonic encoder is configured to process a predefined number Nr of reflections per sound source, and keeps the Nr reflections having the lowest attenuation. In another embodiment of the invention, the virtual sound sources are kept on the basis of a comparison of an acoustic coefficient with a predefined threshold.

The figure 6 represents an example of calculation of early reflections and late reflections, in one embodiment of the invention.

Diagram 600 represents the intensity of several reflections of the sound wave, versus time. The axis 601 represents the intensity of a reflection, and the axis 602 the delay between the emission of the sound wave by the source of the sound wave, and the perception of a reflection by the user. In this example, reflections occurring before a predefined delay 603 are considered early reflections 610, and reflections occurring after delay 603 as late reflections 620. In one embodiment of the invention, early reflections are calculated at using a virtual sound source, for example according to the principle described with reference to the figure 5 .

According to different embodiments of the invention, the late reflections are calculated as follows: a set of Nt secondary sound sources is calculated, for example according to the principle described in figure 5 . The logic for calculating acoustic coefficients and delays, and the position of the virtual sound source of the reflections is configured to keep a number Nr of reflections less than Nt, according to various embodiments described above. In a set of embodiments of the invention, it is further configured to build a list of (Nt - Nr) late reflections, including all non-conserved reflections. This list includes only, for each late reflection, an acoustic coefficient and a delay of the late reflection, but no position of a virtual source.

According to one embodiment of the invention, this list is transmitted by the ambisonic encoder to an ambisonic decoder. The ambisonic decoder is then configured to filter its outputs, for example its output stereo channels, with the acoustic coefficients and the delays of the late reflections, then to add these filtered signals to the output signals. This makes it possible to improve the feeling of immersion in a room or a listening environment, while further limiting the computational complexity of the encoder.

According to another embodiment of the invention, the ambisonic encoder is configured to filter the sound wave with the acoustic coefficients and the delays of the late reflections, and to add the signals obtained uniformly to the set of ambisonic coefficients. This makes it possible to obtain, with a limited computational complexity, an effect representative of multiple reflections in a sound environment. In this embodiment of the invention, as in the previous one, the late reflections have a low intensity and have no direction information from a sound source. They will therefore be perceived by a user as an “echo” of the sound wave, distributed homogeneously in the sound scene, and representative of a listening environment.

The calculation of the acoustic coefficients and delays of the late reflections induces the calculation of many reflections. It is therefore a relatively expensive operation in terms of computational complexity. According to one embodiment of the invention, this calculation is carried out only once, for example on initialization of the sound scene, and the acoustic coefficients and the delays of the late reflections are reused without modification by the ambisonic encoder. This makes it possible to obtain late reflections representative of the listening environment at a lower cost. According to others embodiments of the invention, this calculation is performed iteratively. For example, these acoustic coefficients and delays of the late reflections can be calculated at predefined time intervals, for example every 5 seconds. This makes it possible to permanently conserve acoustic coefficients and delays of late reflections representative of the sound scene, and the relative positions of a source of the sound wave and of the user, while limiting the complexity of calculation linked to the determination. late reflections.

In other embodiments of the invention, the acoustic coefficients and delays of the late reflections are calculated when the position of a source of the sound wave or of the user varies significantly, for example when the difference between the position of the user and a previous position of the user during a calculation of the acoustic coefficients and delays of the late reflections representative of the sound scene is greater than a predefined threshold. This makes it possible to calculate the acoustic coefficients and delays of the late reflections representative of the sound scene only when the position of a source of the sound wave or of the user has varied sufficiently to perceptibly modify the late reflections.

The figure 7 represents a method of encoding a sound wave at a plurality of reflections in a set of embodiments of the invention.

The method 700 comprises a step 710 of frequency transformation of the sound wave.

It then comprises a step 720 of calculating spherical harmonics of the sound wave and of the plurality of reflections from a position of a source of the sound wave and from positions of obstacles to the propagation of waves. sound.

It then comprises a step 730 of filtering, by a plurality of filtering logics in the frequency domain, the spherical harmonics of the plurality of reflections, each filtering logic being parameterized by acoustic coefficients and reflection delays.

It then comprises a step 740 of adding spherical harmonics of the sound wave and of the outputs of the filtering logic.

The above examples demonstrate the ability of an ambisonic encoder according to the invention to calculate ambisonic coefficients of a sound wave at a plurality of reflections. However, they are given only by way of example and in no way limit the scope of the invention, defined in the claims below.

Claims (13)

1. An ambisonic encoder (400) for a sound wave (410) having a plurality of reflections, comprising:

   - a logic (420) for frequency transformation of the sound wave;

   - a logic (430) for calculating spherical harmonics of the sound wave and of the plurality of reflections on the basis of a position of a source of the sound wave and of positions of obstacles to propagation of the sound wave;

   - a plurality (440) of filtering logics in the frequency domain receiving, as input, spherical harmonics of the plurality of reflections, each filtering logic consisting of an attenuation and a delay of a reflection, and being parameterized by an acoustic coefficient and a delay of said reflection;

   - a logic (450) of adding spherical harmonics of the sound wave and outputs from the filtering logics into a set of spherical harmonics representative of both the sound wave and the plurality of reflections in the frequency domain;

   - a multiplication logic of said set of spherical harmonics representative of both the sound wave and the plurality of reflections in the frequency domain by sound intensity values of the wave at the output of the frequency transformation to obtain a set of ambisonic coefficients (B ₀₀, B _1-1, B ₁₀, B ₁₁, ..., B _MM) representative of both the sound wave and the plurality of reflections.
2. The ambisonic encoder according to claim 1, wherein the logic for calculating spherical harmonics of the sound wave is configured to calculate the spherical harmonics of the sound wave and of the plurality of reflections on the basis of a fixed position of the source of the sound wave.
3. The ambisonic encoder according to claim 1, wherein the logic for calculating spherical harmonics of the sound wave is configured to iteratively calculate the spherical harmonics of the sound wave and of the plurality of reflections on the basis of successive positions of the source of the sound wave.
4. The ambisonic encoder according to one of claims 1 to 3, wherein each reflection is characterized by a unique acoustic coefficient.
5. The ambisonic encoder according to one of claims 1 to 3, wherein each reflection is characterized by an acoustic coefficient for each frequency of the frequency sampling.
6. The ambisonic encoder according to one of claims 1 to 5, wherein the reflections are represented by virtual sound sources.
7. The ambisonic encoder according to one of claims 1 to 5, further comprising a logic for calculating the acoustic coefficients, the delays and the position of the virtual sound sources of the reflections, the calculating logic being configured to calculate the acoustic coefficients and the delays of the reflections according to estimates of a difference in the distance traveled by the sound between the position of the source of the sound wave and an estimated position both of a user and of a distance traveled by the sound between the positions of the virtual sound sources of the reflections and the estimated position of the user.
8. The ambisonic encoder according to claim 7, wherein the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is further configured to calculate the acoustic coefficients of the reflections according to at least one acoustic coefficient of at least one obstacle to the propagation of sound waves, off which the sound is reflected.
9. The ambisonic encoder according to one of claims 7 to 8, wherein the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is configured to calculate positions of virtual sound sources of the reflections as inverses of the position of the source of the sound wave with respect to a plane that is tangential to an obstacle to the propagation of sound waves.
10. The ambisonic encoder according to one of claims 1 to 9, wherein the logic for calculating spherical harmonics of the sound wave and of the plurality of reflections is further configured to calculate spherical harmonics of the sound wave and of the plurality of reflections at each output frequency of the frequency transformation circuit, the ambisonic encoder further comprising a logic for calculating binaural coefficients of the sound wave, which logic is configured to calculate binaural coefficients of the sound wave by multiplying, at each output frequency of the circuit for transforming the frequency of the sound wave, the signal of the sound wave by the spherical harmonics of the sound wave and of the plurality of reflections at this frequency.
11. The ambisonic encoder according to one of claims 7 to 9, wherein the logic for calculating the acoustic coefficients, the delays and the positions of the virtual sound sources of the reflections is configured to calculate acoustic coefficients and delays of a plurality of late reflections.
12. A method for ambisonically encoding a sound wave having a plurality of reflections, comprising:

    - performing a frequency transformation (710) of the sound wave;

    - calculating (720) spherical harmonics of the sound wave and of the plurality of reflections on the basis of a position of a source of the sound wave and of positions of obstacles to a propagation of sound waves;

    - filtering (730), by a plurality of logics for filtering in the frequency domain, spherical harmonics of the plurality reflections, each filtering logic consisting of an attenuation and a delay of a reflection, and being parameterized by an acoustic coefficient and a delay of a reflection;

    - adding (740) spherical harmonics of the sound wave and the outputs from the filtering logics, into a set of spherical harmonics representative of both the sound wave and of the plurality of reflections in the frequency domain;

    - multiplying said set of spherical harmonics representative of both the sound wave and the plurality of reflections in the frequency domain by sound intensity values of the wave at the output of the frequency transformation to obtain a set of ambisonic coefficients (B ₀₀, B _1-1, B ₁₀, B ₁₁, ..., B _MM) representative of both the sound wave and the plurality of reflections.
13. A computer program product, comprising computer code instructions stored on a computer readable medium for ambisonic encoding of sound waves with a plurality of reflections, said computer code instructions being configured to:

    - perform a frequency transformation of the sound wave;

    - calculate spherical harmonics of the sound wave and of the plurality of reflections on a basis of a position of a source of the sound wave and of positions of obstacles to a propagation of sound waves;

    - parameterize a plurality of logics for filtering in the frequency domain receiving, as input, spherical harmonics of the plurality of reflections, each filtering logic consisting of an attenuation and a delay of a reflection, and being parameterized by an acoustic coefficient and a delay of said reflection;

    - add spherical harmonics of the sound wave and outputs from the filtering logics, into a set of spherical harmonics representative of both the sound wave and the plurality of reflections in the frequency domain;

    - multiply said set of spherical harmonics representative of both the sound wave and the plurality of reflections in the frequency domain by sound intensity values of the wave at the output of the frequency transformation to obtain a set of ambisonic coefficients (B ₀₀, B _1-1, B ₁₀, B ₁₁, ..., B _MM) representative of both the sound wave and the plurality of reflections;
    when said program runs on the computer.

Images