EP3542545A1

EP3542545A1 - Improved audio headphones device

Info

Publication number: EP3542545A1
Application number: EP17808108.9A
Authority: EP
Inventors: Slim ESSID; Raphael Blouet
Original assignee: Telecom ParisTech; Institut Mines Telecom IMT
Current assignee: BLOUET, RAPHAEL; Institut Mines Telecom IMT
Priority date: 2016-11-21
Filing date: 2017-11-20
Publication date: 2019-09-25
Also published as: TW201820315A; WO2018091856A1; US20200186912A1; FR3059191A1; FR3059191B1

Abstract

The invention relates to a data processing for sound reproduction on a sound reproduction device (DIS), of the headphones or earpiece type, that can be worn by a user in an environment (ENV). The device comprises at least one loudspeaker (HP), at least one microphone (MIC) and a connection to a processing circuit comprising: - an input interface (IN) for receiving signals from at least the microphone, - a processing unit (PROC, MEM) for reading at least an audio content to be reproduced on the loudspeaker, and - an output interface (OUT) for delivering at least the audio signals to be reproduced by the loudspeaker. The processing unit is designed for: a) analysing signals coming from the microphone in order to identify sounds emitted by the environment and corresponding to predetermined target sound classes, b) selecting at least one sound identified according to a user preference criterion, and c) constructing said audio signals to be reproduced by the loudspeaker, by a mix chosen between the audio content and the selected sound.

Description

Advanced headphone device

The invention relates to a portable sound listening device. This may be an audio headset with left and right headphones, or left and right hand-held headphones.

Noise-canceling audio headphones are known, based on a pickup by a microphone array of the user's sound environment. In general, these devices seek to build, in real time, the optimal filter to minimize the contribution of the sound environment in the sound signal perceived by the user. It has recently been proposed a filter of the surrounding noise that may be a function of the type of environment provided by the user himself, who can then select different modes of noise cancellation (office, outside, etc.). The "outside" mode in this case provides a reinjection of the surrounding signal (but at a much lower level than without a filter, and this so as to allow the user to remain aware of its environment).

We also know headphones audio and earphones, selective, allowing a personalized listening of the environment. Recently introduced, these products make it possible to modify the perception of the environment according to two axes:

- the increase in perception (intelligibility of the speech), and

- protection of the hearing aid in a noisy environment.

It can be audio headphones, configurable via a smartphone application. Speech amplification is possible in a noisy environment, where speech is usually located in front of the user.

It can also be audio headphones connected to a smartphone, allowing the user to configure his perception of the sound environment: adjust the volume, add an equalizer or sound effects. We can also mention the interactive headphones and headphones, for augmented reality, to enrich the sound environment (game, pageant) or to support a user activity (virtual coach). Finally, the methods implemented by certain hearing aids to improve the hearing-impaired user's experience propose axes of innovation such as the improvement of the spatial selectivity (according to the direction of the eyes of the user for example) . However, these different existing achievements do not allow:

to analyze and interpret the activity of the user, neither the contents that he consumes, nor the environment (notably the sound stage) in which he is immersed;

to automatically modify the audio rendering according to the result of these analyzes.

Typically, the noise-canceling headphones are based on a sound-only multichannel capture of the user's environment. They seek to reduce overall its contribution to the signal perceived by the user regardless of the nature of the environment, even if it contains potentially interesting information. These devices therefore tend to isolate the user from his environment.

The selective headphones prototypes allow the user to configure his sound environment for example by applying equalization filters or by increasing the intelligibility of speech. These devices make it possible to improve the perception of the environment of the user but do not really make it possible to modify the broadcasted content according to the state of the user or the classes of sounds present in the environment. In this configuration, the user listening to music with a loud volume is always isolated from his environment and the need a device allowing the user to capture the relevant information in his environment is always present.

Certainly, the headphones and interactive earphones can be equipped with sensors to load and broadcast content associated with a place (as part of a tourist visit for example) or an activity (game, sports training). If some devices even have inertial or physiological sensors to monitor the activity of the user and if the dissemination of certain content may then depend on the results of the analysis of the signals from these sensors, the content broadcast does not result from an automatic generation process taking into account the analysis of the sound scene surrounding the user and do not automatically select the components of this environment relevant to the user. Furthermore, the operating modes are static, and do not automatically follow the evolution over time of the sound environment, and even less other evolutionary parameters such as a physiological state for example of the user.

The present invention improves the situation. It proposes for this purpose a method implemented by computer means, data processing for a sound reproduction on a sound reproduction device, headset or earphones, portable by a user in an environment, the device comprising:

- at least one speaker,

- at least one microphone,

a connection to a processing circuit,

the processing circuit comprising:

an input interface for receiving signals from at least the microphone,

a processing unit for reading at least one audio content to be reproduced on the loudspeaker, and an output interface for delivering at least audio signals to be reproduced by the loudspeaker.

In particular, the processing unit is further arranged to implement the steps:

a) analyzing the signals from the microphone to identify sounds emitted by the environment and corresponding to predetermined target sound classes, b) selecting at least one identified sound, according to a user preference criterion, and c) constructing said audio signals to be reproduced by the loudspeaker, by a mix selected between the audio content and the selected sound.

In one possible embodiment, the device comprises a plurality of microphones and the analysis of the signals from the microphones further comprises a sound source separation process in the environment applied to the signals from the microphones.

For example, in step c), the selected sound can be:

- analyzed at least in frequency and duration,

- enhanced by filtering after source separation processing, and mixed with audio content.

In an embodiment where the device comprises at least two loudspeakers and the reproduction of the signals on the loudspeakers applies a 3D sound effect, a sound source position, detected in the environment and emitting a selected sound, can be taken into account. to apply a sound spatialization effect of the source in the mix.

In one embodiment, the device may further comprise a connection to a human-machine interface available to a user for entering preferences for selecting sounds from the environment (in the general sense, as will be seen below) and the criterion The user preference is then determined by learning a history of preferences entered by the user and stored in memory. In an embodiment (alternative or complementary), the device may further comprise a connection to a database of user preferences and the user preference criterion is then determined by analyzing the content of said database.

The device may further include a connection to one or more state sensors of a user of the device, such that the user preference criterion takes into account a current state of the user, thereby contributing to a definition. of the "environment" of the user, in the general sense.

In such an embodiment, the device may comprise a connection to a mobile terminal available to the user of the device, this terminal advantageously comprising one or more state sensors of the user. The processing unit may be further arranged to select a content to be read from among a plurality of contents, depending on the state of the user.

In one embodiment, the predetermined target sound classes may include at least speech sounds whose voiceprints are pre-recorded.

In addition, by way of example, step a) may optionally include at least one of the following operations:

• construction and application of a dynamic filter for cancellation of noise in the signals from the microphone;

Locating and isolating sound sources from the environment by applying a source separation process applied to signals from several microphones, and for example using beamforming, to identify sources of noise. 'interest (for the user of the device) extracting parameters specific to these sources of interest with a view to a subsequent rendering of the sounds picked up and coming from these sources of interest in a spatialized audio mix;

identification of the different classes of sound corresponding to the sources (in different spatial directions) by a classification system (for example by deep neural networks) of known sound classes (speech, music, noise, etc.), and possible identification by d other techniques for classifying the sound stage (for example, sound recognition of a desk, an outdoor street, transportation, etc.).

In addition, by way of example, step c) may optionally include at least one of the following operations:

temporal filtering, spectral filtering and / or spatial filtering (for example Wiener filtering, and / or Duet algorithm), to enhance, from one or more audio streams picked up by a plurality of microphones, a given sound source (in based on the parameters extracted by the aforementioned source separation module);

- 3D audio rendering, for example using HRTF (Head Related Transfer Functions) filtering techniques.

The present invention also relates to a computer program comprising instructions for implementing the above method, when this program is executed by a processor. The invention also relates to a sound reproduction device, of the headset or earphone type, portable by a user in an environment, the device comprising:

- at least one speaker,

- at least one microphone,

a connection to a processing circuit,

the processing circuit comprising:

an input interface for receiving signals from at least the microphone, a processing unit for reading at least one audio content to be reproduced on the loudspeaker, and

an output interface for delivering at least audio signals to be reproduced by the loudspeaker.

The processing unit is further arranged for:

analyzing the signals from the microphone to identify sounds emitted by the environment and corresponding to predetermined classes of target sounds,

selecting at least one identified sound, according to a user preference criterion, and

- Build said audio signals to be reproduced by the speaker, by a mix selected between the audio content and the selected sound.

The invention thus proposes a system including an intelligent audio device, integrating for example a network of sensors, at least one loudspeaker and a terminal (e.g. smartphoned). The originality of this system is to be able to automatically generate, in real time, the "optimal soundtrack" of the user, that is to say the multimedia content best suited to its environment and its state clean.

The user's own state can be defined by: i) a set of preferences (type of music, sound classes of interest, etc.); (ii) his activity (rest, office, sports training, etc.); iii) its physiological states (stress, fatigue, stress, etc.) and / or socio-emotional states (personality, mood, emotions, etc.).

The multimedia content generated may comprise a main audio content (to be broadcast in the headset) and possibly secondary multimedia contents (texts, images, video) that can be broadcast via the smartphoned terminal.

The different content elements include both the elements of the user's content base (music, video, etc., hosted on the terminal or in the cloud), the result of captures made by a sensor network that includes the system and synthetic elements generated by the system (notifications, "jingles" sound or text, noise comfort, etc.).

Thus, the system can automatically analyze the user's environment and predict the components potentially of interest to the user in order to restore them in an increased and controlled manner, superimposing them optimally on the contents consumed by the user ( typically the music he listens to).

The actual rendering of the contents takes into account the nature of the contents and components extracted from the environment (as well as the user's own state in a more sophisticated embodiment). The sound flow diffused in the helmet no longer comes from two competing sources:

- a main source (music or radio program or other), and

- a disturbing source (ambient noise), but a set of information flows whose relative contributions are adjusted according to their relevance. Thus, a message broadcast in the enclosure of a station will be restored so that it is well perceived by the user while he listens to music at a high level, while reducing the ambient noise irrelevant for the user. This possibility is offered by the addition of an intelligent processing module including algorithms for separating sources and classification of sound scenes. The direct application advantage is, on the one hand, to reconnect the user with his environment or to warn him if a class of targeted sounds is detected, and on the other hand to automatically generate content adapted at each moment to the expectations of the user. user through a recommendation engine supporting the various content elements, supra.

It should be remembered that state-of-the-art devices do not make it possible to automatically identify each class of sound present in the user's environment in order to associate with each of them a treatment that meets the expectations of the user. the user (for example a highlighting of a sound, or on the contrary a reduction, the generation of an alert), according to its identification in the environment. The state of the art does not use analysis of the sound stage, nor the state of the user or his activity to calculate the sound reproduction.

Other advantages and characteristics of the invention will appear on reading the detailed description of examples of embodiment below and on examining the appended drawings in which:

FIG. 1 illustrates a device according to the invention, in a first embodiment,

FIG. 2 illustrates a device according to the invention, in a second embodiment, here connected to a mobile terminal,

FIG. 3 illustrates the steps of a method according to one embodiment of the invention, and FIG. 4 specifies steps of the method of FIG. 3, according to a particular embodiment.

With reference to FIG. 1, a sound reproduction device DIS (of the headset or earpiece type), worn for example by a user in an environment ENV, comprises at least:

one (or two, in the example represented) speakers HP, at least one sensor, for example a microphone MIC (or a row of microphones in the example shown to capture a directivity sounds from the environment) and - a connection to a processing circuit.

The processing circuit can be integrated directly into the headphones and housed in a loudspeaker enclosure (as illustrated in FIG. 1), or can, in the variant illustrated in FIG. 2, be implemented in a TER terminal of the user, by example a smartphone-type mobile terminal, or be distributed between several terminals of the user (a smartphone, and a connected object possibly including other sensors). In this variant, the connection between the headset (or headsets) and the dedicated processing circuit of the terminal is performed by a USB connection or short-range radio frequency (for example by Bluetooth or other) and the headset (or headsets) is equipped with a transmitter / receiver BT1, communicating with a transceiver BT2 that includes the terminal TER. A hybrid solution in which the processing circuit is distributed between the speaker of the headset and a terminal is also possible.

In either of the embodiments above, the processing circuit comprises:

an input interface IN, for receiving signals coming from at least the microphone MIC, a processing unit typically comprising a processor PROC and a memory MEM, for interpreting, with respect to the environment ENV, the signals coming from the microphone by learning (for example by classification, or by "matching" type "finger printing" for example),

an output interface OUT for delivering at least audio signals that are functions of the environment and to be reproduced by the loudspeaker.

The memory MEM can store instructions of a computer program within the meaning of the present invention, and possibly temporary data (calculation or otherwise), as well as durable data, such as the user's preferences, or even model definition data or other, as will be discussed later. The input interface IN is, in a sophisticated embodiment, connected to a network of microphones, as well as to an inertial sensor (provided on the headphones or in the terminal) and the definition of user preferences. The user preference data may be stored locally in the MEM memory, as indicated above. As a variant, they can be stored, possibly with other data, in a remote database DB accessible via a communication via a local or extended NW network. An LP communication module with such a network suitable for this purpose can be provided in the headset or in the TER terminal.

Advantageously, a man / machine interface can allow the user to define and update his preferences. In the embodiment of FIG. 2, where the device DIS is paired with the terminal TER, the man / machine interface can simply correspond to a touch screen of the smartphone TER for example. Otherwise, it can be provided such an interface directly on the headset.

In the embodiment of FIG. 2, however, it is advantageously possible to take advantage of the presence of additional sensors in the TER terminal to enrich the definition of the environment of the user, in the general sense. These additional sensors may be physiological sensors specific to the user (measurement of electroencephalogram, heart rate measurement, pedometer, etc.) or any other sensor to improve the knowledge of the environment / current state of the user. In addition, this definition can directly include the notification by the user himself of his activity, his own condition and his environment.

The definition of the environment can take into account:

all the accessible contents and a history of the contents consulted (music, videos, radio broadcasts, etc.),

metadata (for example the genre, the listening occurrences per piece) associated with the user's music library may also be associated; - Moreover, the browsing history and applications of his smartphone; the history of its consumption of streaming content (via a service provider) or locally;

preferences and current activity of its connections on social networks.

Thus, the input interface may, in the general sense, be connected to a set of sensors, and also include connection modules (including the LP interface) for characterizing the user's environment, but also habits and preferences (history of content consumption, streaming activities and / or social networks).

With reference to FIG. 3, the treatment operated by the aforementioned processing unit, monitoring the environment and possibly the state of the user to characterize the relevant information that can be restored to the stream is described below. multimedia output. In one embodiment, this monitoring is implemented by automatic extraction, via signal processing and artificial intelligence modules, in particular machine learning modules (represented by step S7 in FIG. important for creating the output media stream. These parameters, denoted PI, P2, ..., in the figures can typically be environment parameters which must be taken into account for the reproduction on loudspeakers. For example, if a sound picked up in the environment is identified as a speech signal to be played back:

a first set of parameters may be coefficients of an optimal filter (Wiener filter type) making it possible to enhance the speech signal to increase its intelligibility;

a second parameter is the directivity of the sound captured in the environment and to be rendered for example by means of a binaural rendering (rendering technique using transfer functions of HRTF type);

- etc. It will thus be understood that these parameters PI, P2, are to be interpreted as "descriptors" of the environment and the general state of the user in the general sense, which feed a program for generating the "optimal soundtrack" for this user. This soundtrack is obtained by composition of its contents, elements of the environment and synthetic elements.

During the first step S 1, the processing unit requests the input interface to collect the signals from the microphone or microphone array MIC that the DIS carries. Of course, other sensors (inertia, or others) in the terminal TER in step S2, or elsewhere in step S3 (connected heart rate sensors, EEG, etc.), can communicate their signals to the processing unit. On the other hand, information data other than captured signals (user's preferences in step S5, and / or the consumption history of content and connections to the social networks in step S6) can be transmitted. by the memory MEM and / or by the database BD to the processing unit.

In step S4, all these data and signals specific to the environment and the state of the user (hereinafter referred to generically as "environment") are collected and interpreted by the implementation, at the step S7, a computer module for decoding the environment by artificial intelligence. For this purpose, this decoding module can use a learning base which can, for example, be remote and requested in step S8 via the network NW (and the communication interface LP), in order to extract parameters relevant PI, P2, P3, at step S9 which model the environment in general.

As further detailed with reference to FIG. 4, from these parameters in particular, the sound scene to be rendered is generated in step S 10 and transmitted in the form of audio signals to the loudspeakers HP at step SU. This sound scene may possibly be accompanied by graphic information, for example metadata, to be displayed on the screen of the terminal TER in step S12.

Thus, an analysis of the environmental signals is carried out, with: an identification of the environment with a view to estimating prediction models making it possible to characterize the user's environment and his or her own state (these models being used with a recommendation engine as will be seen below with reference to the figure 4), and

- A fine acoustic analysis to generate more precise parameters and used to manipulate the audio content to be restored (separation / enhancement of particular sound sources, sound effects, mixing, spatialization, or other).

The identification of the environment makes it possible to characterize, by automatic learning, the environment / state pair of the user. It is mainly:

detect whether certain classes of target sounds, among several prerecorded classes, are present in the environment of the user and determine, if necessary, their direction of provenance. Initially, the classes of its target may be defined, one by one, by the user via his terminal or by using predefined operating modes;

determine the activity of the user: rest, at the office, active in a gym, or other;

to determine the emotional and physiological state of the user (for example "in shape", according to a pedometer, or "stressed" according to his EEG);

describe the content they consume using content analysis techniques (computer-based hearing and vision techniques, and natural language processing). The fine acoustic analysis makes it possible to calculate the acoustic parameters which are used for the audio reproduction (for example in 3D restitution).

Referring now to FIG. 4, in step S17, a recommendation engine is used to receive the descriptors of the "environment", in particular the classes of identified sound events (parameters PI, P2, etc.). , and provide on this basis a recommendation model (or a combination of models) at step S 19. At this indeed, the recommendation engine may use the characterization of the user's contents and their similarity to external contents as well as user preferences, which have been recorded in a learning base in step S 15, and / or the standard preferences of other users in step S18. The user can also intervene at this step with his terminal to enter a preference at step S24, for example with respect to a content or a list of contents. to play.

From all these recommendations, a recommendation model is chosen that is relevant to the environment and the state of the user (for example, in the group of rhythmic music, in a situation of movement of the apparently user in a gym). A composition engine is then implemented in step S20, which combines the parameters PI, P2 ..., with the recommendation model, to develop a composition program in step S21. This is a routine that suggests for example:

a specific type of content to be searched for in the contents of the user, taking into account its own state (for example its activity) and certain types of sounds of the external environment identified in the parameters PI, P2,. ..,

- To mix to the content, according to a sound level and a spatial rendering (3D audio) that has been defined by the composition engine.

The synthesis engine, strictly speaking, of the audible signal intervenes in step S22, to elaborate the signals to be restored to the steps SI 1 and S 12, from:

the contents of the user (from step S25 (as a substep of step S6), of course, one of the contents having been selected in step S21 by the composition engine,

sound signals picked up in the environment (IF, possibly parameters PI, P2, ... in the case of a synthesis of the sounds of the environment to be rendered), and

other sounds, possibly synthetic, of notifications (beep, bell, or other), able to announce an external event and to mix the content to be restored (selected in step S21 from step S 16), possibly with a 3D rendering defined in step S23.

Thus, the stream generated is adapted to the expectations of the user and optimized according to the context of its distribution, according to three main steps in a particular embodiment:

the use of a recommendation engine for filtering and selecting in real time the content elements to be mixed for the sound reproduction (and possibly also visual) of a multimedia flow (called "controlled reality");

the use of a media composition engine which programs the temporal, frequency and spatial arrangement of the content elements, with respective sound levels also defined;

the use of a synthesis engine generating the signals of the sound (and possibly visual) rendering, possibly with sound spatialization, according to the program established by the composition engine.

The generated multimedia stream includes at least audio signals but potentially textual, haptic and / or visual notifications. The audio signals include a mix:

a content selected in the user's content database (music, video, etc.), entered as a preference by the user in step S24, or recommended directly by the recommendation engine according to the state of the user and the environment,

with possibly

sounds picked up by the MIC sensor network, selected in the sound environment (thus filtered), enhanced (for example by source separation techniques) and processed so that they are of frequency texture, intensity and spatialisation, adjusted to be injected into the mix in a timely fashion, and

synthetic elements recovered from a base at step S 16, for example sound / text jingles sounds, comfort noise, etc.). The recommendation engine is based on:

the user's preferences obtained explicitly through a form of questioning, or implicitly by exploiting the result of the decoding of his own state,

collaborative filtering techniques and social graphs, exploiting the models of several users at once (step S 18),

the description of the user's content and their similarity, in order to build models for deciding which content elements should be played to the user.

The models are updated continuously over time to adapt to the evolution of the user.

The composition engine plans:

- the time at which each piece of content must be played, including the order in which the user's contents are presented (for example, the order of the music in a playlist), and the external moments or sounds or the notifications are broadcast: in real time or delayed (for example between two pieces of a playlist) so as not to disturb the listening or the current activity of the user at an inconvenient time;

- the spatial position (for 3D rendering) of each piece of content;

- the different audio effects (gain, filtering, equalization, dynamic compression, echo or reverb ("reverb"), slowdown / temporal acceleration, transposition ...) that must be applied to each element of content.

Planning is based on templates and rules built from decoding the user's environment and their own state. For example, the spatial position of a sound event captured by the pickups and the gain level associated with it depend on the result of the sound source localization detection performed by the decoding of the environment in step S7 of the figure 3. The synthesis engine is based on signal processing techniques, natural languages and images, respectively for the synthesis of audio, textual and visual outputs (images or videos), and jointly for the generation of multimedia outputs, for example video.

In the case of the synthesis of the audio output, temporal, spectral and / or spatial filtering techniques can be exploited. For example, the synthesis is first performed locally on short time windows and the signal is reconstructed by addition-overlap before being transmitted to at least two speakers (one for each ear). Different gains (power levels) and audio effects are applied to the different content elements as provided by the composition engine. In a particular embodiment, the processing applied by windows may include a filtering (for example of Wiener) making it possible to enhance, from one or more of the audio streams captured, a particular sound source (as provided by the composition engine). . In a particular embodiment, the processing may include 3D audio rendering, possibly using HRTF filtering techniques (HRTF transfer functions for "Head Related Transfer Functions").

In a first example illustrating a minimal implementation, the description of the environment of the user is limited to his sound environment; the user's own state is limited to his preferences: class of his target, notifications he wishes to receive, these preferences being defined by the user using his terminal; the device (possibly in cooperation with the terminal) is equipped with inertial sensors (accelerometer, gyroscope and magnetometer); playback parameters are automatically changed when a target sound class is detected in the user's environment; - messages of short duration can be recorded; notifications can be sent to the user to warn him of the detection of an event of interest.

The captured signals are analyzed to determine:

- the sound classes present in the user's environment and the directions from which they come, with, for this purpose:

a detection of the directions of higher sound energies by analyzing the contents in each of these directions independently,

- an overall determination for each direction of the contribution of each sound class (eg using a source separation technique),

- Model parameters describing the user's environment and those parameters feeding the recommendation engine.

In a second example illustrating a more sophisticated implementation, a set of sensors including a network of microphones, a video camera, a pedometer, inertial sensors (accelerometers, gyroscopes, magnetometers), physiological sensors can capture the visual and sound environment of the user (microphones and camera), the data characterizing his movement (inertial sensors, pedometer) and his physiological parameters (EEG, ECG, EMG, electrodermal) as well as all the contents he is consulting (music , radio broadcasts, videos, browsing history and applications of his smartphone). Then, the different flows are analyzed to extract the information related to the user's activity, his mood, his state of fatigue and his environment (for example running on carpets riding in a gym, in a good mood and in a state of low fatigue). A musical stream adapted to the environment and to the user's own state can be generated (for example a playlist of which each piece is selected according to its musical tastes, its stride and its state of fatigue). While all sound sources are canceled in the user's headset, the voice of a coach ("coach") near the user, when it is identified (voice record previously recorded), is mixed at stream and rendered spatially using binaural rendering techniques (by HRTF for example).

Claims

1. Method implemented by computer means for processing data for sound reproduction on a sound reproduction device, of the headset or earpiece type, portable by a user in an environment, the device comprising: at least one loudspeaker; speaker,

- at least one microphone,

a connection to a processing circuit,

the processing circuit comprising:

an input interface for receiving signals originating at least from the microphone; a processing unit for reading at least one audio content to be reproduced on the loudspeaker, and

an output interface for delivering at least audio signals to be reproduced by the loudspeaker,

characterized in that the processing unit is further arranged to implement the steps:

a) analyzing the signals from the microphone to identify sounds emitted by the environment and corresponding to predetermined target sound classes, b) selecting at least one identified sound, according to a user preference criterion, and c) constructing said audio signals to be reproduced by the loudspeaker, by a mix selected between the audio content and the selected sound,

and in that the device comprising a plurality of microphones, the analysis of the signals from the microphones further comprises a sound source separation process in the environment applied to the signals from the microphones.

2. Method according to claim 1, characterized in that, in step c), the selected sound is:

- analyzed at least in frequency and duration,

3. Method according to one of the preceding claims, characterized in that, the device comprising at least two speakers and the return of the signals on the speakers applying a 3D sound effect, a sound source position, detected in the environment and emitting a selected sound, is taken into account to apply a sound spatialization effect of the source in the mix.

4. Method according to one of the preceding claims, characterized in that the device further comprises a connection to a man-machine interface available to a user for entering preferences for selecting sounds from the environment, and in that the user preference criterion is determined by learning a history of preferences entered by the user and stored in memory.

5. Method according to one of the preceding claims, characterized in that the device further comprises a connection to a database of user preferences and the user preference criterion is determined by analysis of the content of said database. data.

6. Method according to one of the preceding claims, characterized in that the device further comprises a connection to one or more state sensors of a user of the device, and in that the user preference criterion takes into account a current state of the user.

7. Method according to claim 6, characterized in that the device comprises a connection to a mobile terminal available to the user of the device, the terminal comprising one or more state sensors of the user.

8. Method according to one of claims 6 and 7, characterized in that the processing unit is further arranged to select a content to be read from among a plurality of contents, depending on the state of the user.

9. Method according to one of the preceding claims, characterized in that the classes of target sounds, predetermined, comprise at least speech sounds, prerecorded voice prints.

10. Computer program characterized in that it comprises instructions for the implementation of the method according to one of claims 1 to 9, when the program is executed by a processor.

11. A sound reproduction device, of the headset or earphone type, portable by a user in an environment, the device comprising:

- at least one speaker,

- at least one microphone,

a connection to a processing circuit,

the processing circuit comprising:

an input interface for receiving signals from at least the microphone,

a processing unit for reading at least one audio content to be reproduced on the loudspeaker, and

characterized in that the processing unit is further arranged for:

constructing said audio signals to be reproduced by the loudspeaker, by a mix chosen between the audio content and the selected sound,