EP4287648A1 - Electronic device and associated processing method, acoustic apparatus and computer program - Google Patents

Abstract

This electronic processing device (20) for an acoustic device (10) comprising a first aerial microphone (12) and a second osteophonic microphone (14), is configured to be connected to the first and second microphones (12,14), to receive as input to the first, and respectively second, analog signals coming from the first, and respectively second, microphones (12,14) and to deliver a corrected signal at the output. The processing device (20) comprises: - a hybridization module ( 30) configured to calculate a hybrid signal from the first and second analog signals; - an estimation module (32) configured to estimate noise in the hybrid signal; - a noise reduction module (34) configured to calculate the signal corrected by applying a generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise.

Description

The present invention relates to an electronic processing device for an acoustic apparatus.

The invention also relates to an acoustic apparatus comprising a first microphone comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from a user's vocal cords and of transforming said acoustic waves into a first analog signal; a second microphone comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal; and such an electronic processing device connected to the first and second microphones, the processing device being configured to receive the first and second analog signals as input, then to output a corrected signal.

The electronic processing device includes a hybridization module configured to calculate a hybrid signal from the first and second analog signals.

The invention also relates to a processing method implemented by such an electronic processing device; as well as a computer program comprising software instructions which, when executed by a computer, implement such a processing method.

We know of the document FR 3 019 422 B1 an acoustic device of the aforementioned type. The acoustic apparatus includes the first microphone with such an electroacoustic transducer, also called an aerial transducer; the second microphone with such a bone mechanical excitation transducer, also called solid-body transducer; means for calculating a corrected electrical signal as a function of the first electrical signal and the second electrical signal, the corrected electrical signal being suitable for being delivered at the output of the acoustic device; and a noise reduction device connected to the output of the electroacoustic transducer to reduce noise in the first electrical signal; the calculation means being connected, on the one hand, to the output of the noise reduction device, and on the other hand, to the output of the mechanical bone excitation transducer.

However, with such an acoustic device, the noise reduction is not always optimal, and there is sometimes relatively high background noise in the signal delivered at the output of the acoustic device.

The aim of the invention is then to propose an electronic processing device, and an associated processing method, making it possible to further improve the reduction of noise in the signal delivered at the output of the acoustic device, that is to say say to reduce the presence of noise in said signal.

• l'appareil acoustique comprenant un premier microphone comportant un transducteur électroacoustique apte à recevoir des ondes sonores acoustiques d'un signal sonore issu de cordes vocales d'un utilisateur et à transformer lesdites ondes acoustiques en un premier signal analogique ; et un deuxième microphone comportant un transducteur à excitation mécanique osseuse apte à recevoir par conduction osseuse des oscillations vibratoires dudit signal sonore et à transformer lesdites oscillations vibratoires en un deuxième signal analogique,
• le dispositif électronique de traitement étant configuré pour être connecté aux premier et deuxième microphones, pour recevoir en entrée les premier et deuxième signaux analogiques et pour délivrer en sortie un signal corrigé,
• le dispositif électronique de traitement comprenant :
  • un module d'hybridation configuré pour calculer un signal hybride à partir des premier et deuxième signaux analogiques ;
  • un module d'estimation connecté au module d'hybridation et configuré pour estimer un bruit dans le signal hybride ; et
  • un module de réduction de bruit connecté au module d'hybridation et au module d'estimation, le module de réduction de bruit étant configuré pour calculer le signal corrigé en appliquant un algorithme de soustraction spectrale généralisée au signal hybride et en fonction du bruit estimé.

To this end, the subject of the invention is an electronic processing device for an acoustic device,

• the acoustic apparatus comprising a first microphone comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from a user's vocal cords and of transforming said acoustic waves into a first analog signal; and a second microphone comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal,
• the electronic processing device being configured to be connected to the first and second microphones, to receive the first and second analog signals as input and to output a corrected signal,
• the electronic processing device comprising:
  • a hybridization module configured to calculate a hybrid signal from the first and second analog signals;
  • an estimation module connected to the hybridization module and configured to estimate noise in the hybrid signal; And
  • a noise reduction module connected to the hybridization module and to the estimation module, the noise reduction module being configured to calculate the corrected signal by applying a generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise.

With the electronic processing device according to the invention, the fact of estimating the noise in the hybrid signal calculated from the first and second analog signals, that is to say in the hybrid signal obtained from the signals resulting from 'on the one hand the electroacoustic, or aerial, transducer, and on the other hand the transducer with mechanical bone excitation, also called osteophonic, or even solidian, transducer, makes it possible to have a more precise estimate of the noise, then to obtain - via the noise reduction module - a better signal corrected by applying the generalized spectral subtraction algorithm to the hybrid signal and based on the noise thus estimated.

Preferably, the hybrid signal comprises several successive sections, each section corresponding to the hybrid signal during a time period, and the device processing further comprises a voice activity detection module capable of determining whether each section of the hybrid signal includes a presence of voice or not, the estimation module then being configured to estimate the noise in the hybrid signal only from every voiceless stretch.

The presence or absence of voice is preferably still determined from the second signal from the osteophonic transducer, the presence or absence of voice being better detectable in a signal coming from an osteophonic microphone, rather than in a signal coming from from an overhead microphone.

• le signal hybride comporte plusieurs tronçons successifs, et le dispositif comprend en outre un module de détection d'activité vocale connecté au module d'hybridation et configuré pour déterminer une présence de voix ou une absence de voix dans chaque tronçon du signal hybride ; le module d'estimation étant alors configuré pour estimer le bruit dans le signal hybride en fonction de chaque tronçon avec une absence déterminée de voix ;
• le module de détection d'activité vocale est configuré pour déterminer la présence de voix ou l'absence de voix à partir du deuxième signal issu du transducteur à excitation mécanique osseuse ;
  le module de détection d'activité vocale étant de préférence configuré pour déterminer la présence de voix ou l'absence de voix uniquement à partir du deuxième signal, sans prise en compte du premier signal ;
• le deuxième signal comporte plusieurs tronçons successifs, et le module de détection d'activité vocale est configuré pour calculer une valeur RMS pour chaque tronçon du deuxième signal, puis pour déterminer la présence de voix ou l'absence de voix en fonction de valeur(s) RMS respective(s) ;
• le module de détection d'activité vocale est configuré pour déterminer la présence de voix ou l'absence de voix en fonction d'une valeur moyenne de M dernière(s) valeur(s) RMS calculée(s) et/ou d'une variation de valeur RMS entre une valeur RMS courante et une valeur RMS précédente, M étant un nombre entier supérieur ou égal à 1 ;
  le module de détection d'activité vocale étant de préférence configuré pour déterminer la présence de voix si ladite valeur moyenne est supérieure ou égale à un seuil prédéfini de moyenne ou si ladite variation de valeur RMS est supérieure ou égale à un seuil prédéfini de variation ;
• le module d'hybridation est configuré pour convertir le premier signal analogique en un premier signal numérique, au fur et à mesure de la réception du premier signal analogique, et pour générer des premiers tronçons successifs à partir du premier signal numérique, chaque nouveau premier tronçon généré comportant des échantillons d'un premier tronçon précédent et de nouveaux échantillons du premier signal numérique ; et
  • le module d'hybridation est configuré pour convertir le deuxième signal analogique en un deuxième signal numérique, au fur et à mesure de la réception du deuxième signal analogique, et pour générer des deuxièmes tronçons successifs à partir du deuxième signal numérique, chaque nouveau deuxième tronçon généré comportant des échantillons d'un deuxième tronçon précédent et de nouveaux échantillons du deuxième signal numérique ;
  • des tronçons hybrides du signal hybride étant alors calculés au fur et à mesure à partir des premiers et deuxièmes tronçons générés ; le signal corrigé étant ensuite calculé à partir desdits tronçons hybrides ;
• le module d'hybridation est configuré pour obtenir un premier signal filtré en appliquant au premier signal un premier filtre associé à une première plage de fréquences ; pour obtenir un deuxième signal filtré en appliquant au deuxième signal un deuxième filtre associé à une deuxième plage de fréquences ; puis pour calculer le signal hybride en sommant le premier signal filtré et le deuxième signal filtré, la deuxième plage de fréquences étant distincte de la première plage de fréquences ;
  • la première plage de fréquences comportant de préférence des fréquences supérieures à celles de la deuxième plage de fréquences ;
  • les première et deuxième plages de fréquences étant de préférence encore disjointes.

According to other advantageous aspects of the invention, the electronic processing device comprises one or more of the following characteristics, taken in isolation or in all technically possible combinations:

• the hybrid signal comprises several successive sections, and the device further comprises a voice activity detection module connected to the hybridization module and configured to determine a presence of voice or an absence of voice in each section of the hybrid signal; the estimation module then being configured to estimate the noise in the hybrid signal as a function of each section with a determined absence of voice;
• the voice activity detection module is configured to determine the presence of voice or the absence of voice from the second signal from the bone mechanical excitation transducer;
  the voice activity detection module being preferably configured to determine the presence of voice or the absence of voice solely from the second signal, without taking into account the first signal;
• the second signal comprises several successive sections, and the voice activity detection module is configured to calculate an RMS value for each section of the second signal, then to determine the presence of voice or the absence of voice as a function of value(s ) respective RMS(s);
• the voice activity detection module is configured to determine the presence of voices or the absence of voices based on an average value of M last calculated RMS value(s) and/or a variation in RMS value between a current RMS value and a previous RMS value, M being an integer greater than or equal to 1;
  the voice activity detection module being preferably configured to determine the presence of voices if said average value is greater than or equal to a predefined average threshold or if said RMS value variation is greater than or equal to a predefined variation threshold;
• the hybridization module is configured to convert the first analog signal into a first digital signal, as the first analog signal is received, and to generate successive first sections from the first digital signal, each new first section generated comprising samples of a first previous section and new samples of the first digital signal; And
  • the hybridization module is configured to convert the second analog signal into a second digital signal, as the second analog signal is received, and to generate successive second sections from the second digital signal, each new second section generated comprising samples of a second previous section and new samples of the second digital signal;
  • hybrid sections of the hybrid signal then being calculated progressively from the first and second sections generated; the corrected signal then being calculated from said hybrid sections;
• the hybridization module is configured to obtain a first filtered signal by applying to the first signal a first filter associated with a first frequency range; to obtain a second filtered signal by applying to the second signal a second filter associated with a second frequency range; then to calculate the hybrid signal by summing the first filtered signal and the second filtered signal, the second frequency range being distinct from the first frequency range;
  • the first frequency range preferably comprising frequencies higher than those of the second frequency range;
  • the first and second frequency ranges being preferably still disjoint.

• un premier microphone comportant un transducteur électroacoustique apte à recevoir des ondes sonores acoustiques d'un signal sonore issu de cordes vocales d'un utilisateur et à transformer lesdites ondes acoustiques en un premier signal analogique ;
• un deuxième microphone comportant un transducteur à excitation mécanique osseuse apte à recevoir par conduction osseuse des oscillations vibratoires dudit signal sonore et à transformer lesdites oscillations vibratoires en un deuxième signal analogique ;
• un dispositif électronique de traitement connecté aux premier et deuxième microphones, le dispositif électronique de traitement étant configuré pour recevoir en entrée les premier et deuxième signaux analogiques, puis pour délivrer en sortie un signal corrigé ; le dispositif électronique de traitement étant tel que défini ci-dessus.

The invention also relates to an acoustic device comprising:

• a first microphone comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from the vocal cords of a user and of transforming said acoustic waves into a first analog signal;
• a second microphone comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal;
• an electronic processing device connected to the first and second microphones, the electronic processing device being configured to receive the first and second analog signals as input, then to output a corrected signal; the electronic processing device being as defined above.

According to another advantageous aspect of the invention, the acoustic device further comprises two lateral acoustic modules resting on the lateral sides of the skull and capable of transmitting a sound signal to the auditory nerve.

The invention also relates to head equipment for an operator comprising a protective helmet, and an acoustic device as defined above.

• une étape d'hybridation comportant le calcul d'un signal hybride à partir des premier et deuxième signaux analogiques ;
• une étape d'estimation d'un bruit dans le signal hybride ; et
• une étape de réduction de bruit comportant le calcul du signal corrigé en appliquant un algorithme de soustraction spectrale généralisée au signal hybride et en fonction du bruit estimé.

The invention also relates to a processing method, the method being implemented by an electronic processing device connected to first and second microphones, the first microphone comprising an electroacoustic transducer capable of receiving acoustic sound waves of a signal sound coming from a user's vocal cords and transforming said acoustic waves into a first analog signal; and the second microphone comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal, the electronic processing device being configured to receive as input the first and second analog signals and to output a corrected signal,
the treatment process comprising:

• a hybridization step comprising the calculation of a hybrid signal from the first and second analog signals;
• a step of estimating noise in the hybrid signal; And
• a noise reduction step comprising the calculation of the corrected signal by applying a generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise.

The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement a processing method as defined above.

• la figure 1 est une vue d'ensemble en perspective d'un appareil acoustique selon l'invention, l'appareil acoustique comprenant un premier microphone aérien, un deuxième microphone ostéophonique, et un dispositif électronique de traitement à délivrer un signal électrique corrigé à partir des signaux électriques issus des premier et deuxième microphones ;
• la figure 2 est une représentation schématique sous forme d'un synoptique du dispositif de traitement de la figure 1, connecté au premier microphone aérien et au deuxième microphone ostéophonique ;
• la figure 3 est une représentation schématique d'une génération de tronçons chevauchés, effectuée par le dispositif de traitement de la figure 1 ;
• la figure 4 est un organigramme d'un procédé de traitement selon l'invention, le procédé étant mis en oeuvre par le dispositif de traitement de la figure 1 ;
• la figure 5 est une vue représentant, en partie supérieure, un signal de voix bruité enregistré par un microphone aérien de l'état de la technique ; et en partie inférieure, un signal hybride obtenu avec les premier et deuxième microphones, et après réduction de bruit via le dispositif de traitement de la figure 1 ;
• la figure 6 est une vue avec plusieurs courbes illustrant une détection d'activité vocale de l'état de la technique, via un microphone aérien et pour un seuil de détection bas ;
• la figure 7 est une vue analogue à celle de la figure 6, pour un seuil de détection plus élevé ; et
• la figure 8 est une vue analogue à celles des figures 6 et 7, illustrant une détection d'activité vocale selon l'invention, via un microphone ostéophonique.

These characteristics and advantages of the invention will appear more clearly on reading the description which follows, given solely by way of non-limiting example, and made with reference to the appended drawings, in which:

• there figure 1 is an overall perspective view of an acoustic device according to the invention, the acoustic device comprising a first aerial microphone, a second osteophonic microphone, and an electronic processing device for delivering an electrical signal corrected from the electrical signals from the first and second microphones;
• there figure 2 is a schematic representation in the form of a synoptic of the device for processing the figure 1 , connected to the first aerial microphone and the second osteophonic microphone;
• there Figure 3 is a schematic representation of a generation of overlapped sections, carried out by the processing device of the figure 1 ;
• there figure 4 is a flowchart of a processing method according to the invention, the method being implemented by the processing device of the figure 1 ;
• there figure 5 is a view representing, in the upper part, a noisy voice signal recorded by an aerial microphone of the state of the art; and in the lower part, a hybrid signal obtained with the first and second microphones, and after noise reduction via the noise processing device. figure 1 ;
• there Figure 6 is a view with several curves illustrating voice activity detection of the state of the art, via an aerial microphone and for a low detection threshold;
• there Figure 7 is a view analogous to that of the Figure 6 , for a higher detection threshold; And
• there figure 8 is a view analogous to those of figures 6 And 7 , illustrating detection of vocal activity according to the invention, via an osteophonic microphone.

In the remainder of the description, the expression “substantially equal to” defines a relationship of equality of plus or minus 20%, more preferably more or less 10%, more preferably more or less 5%. .

On the figure 1 , an acoustic device 10 comprises a first microphone 12, also called an aerial microphone, capable of receiving acoustic sound waves and transforming them into a first electrical signal, such as a first analog signal, and a second microphone 14, also called microphone osteophonic or even solid-body microphone, capable of receiving vibratory oscillations by bone conduction and transforming them into a second electrical signal, such as a second analog signal.

The acoustic apparatus 10 comprises a protective housing 18 and a processing device 20 disposed within the protective housing 18, the processing device 20 being connected to the first microphone 12 and the second microphone 14, and configured to receive in input the first and second analog signals and output a corrected signal in which the noise has been reduced.

In addition, the acoustic device 10 also comprises two side acoustic modules 22, an upper arch 24, a rear arch 26 for connecting the acoustic modules and a connection cable 27, the connection cable 27 being equipped at its end with a connector, not shown. The side acoustic modules 22, the upper arch 24, the rear arch 26 and the connection cable 27 are known per se, for example from the document FR 3 019 422 B1 .

The first microphone 12 is known, for example from the document FR 3 019 422 B1 , and comprises an electroacoustic transducer, not shown, capable of receiving waves acoustic sound waves of a sound signal coming from the vocal cords and transforming said acoustic waves into the first electrical signal. The first microphone 12 is connected to the input of the processing device 20.

The second microphone 14 is also known, for example from the document FR 3 019 422 B1 , and comprises a transducer with mechanical bone excitation, not shown, capable of receiving by bone conduction, in particular through a corresponding bone of the skull, the vibration waves of the sound signal coming from the vocal cords of the user and of transforming it into the second electrical signal. The bone mechanical excitation transducer is also called osteophonic transducer, or solid-body transducer. The second microphone 14 is also connected to the input of the processing device 20.

In the example of the figure 1 , the first microphone 12 and the second microphone 14 are not arranged in the protective housing 18, but are arranged in an additional housing 28, the additional housing 28 being connected to one of the two acoustic modules 22 by two connecting arms 29. The electroacoustic transducer and the mechanical bone excitation transducer are then each arranged in the additional housing 28. This additional housing 28 is preferably intended to be applied in contact with the right side of the user's skull, and is then preferably connected to the right acoustic module 22.

Alternatively, as illustrated in the example of Figure 13 of the document FR 3 019 422 B1 , the second microphone 14 is not placed in the protective housing 18, but is arranged in another additional housing, the other additional housing being connected to one of the two acoustic modules 22 by two connecting arms. The bone mechanical excitation transducer of the second microphone is then placed in the other additional housing. This other additional housing is preferably intended to be applied in contact with the right side of the user's skull, and is then preferably connected to the right acoustic module 22.

As a further variation, as illustrated in the example of the figure 1 of the document FR 3 019 422 B1 , the first microphone 12 comprises a protuberance, for example made integrally with the protective housing 18. According to this variant, the second microphone 14, in particular its bone mechanical excitation transducer, is arranged inside the protective housing 18 .

The electronic processing device 20 comprises a hybridization module 30 connected to the first microphone 12 and the second microphone 14; an estimation module 32 connected to the hybridization module 30; and a noise reduction module 34 connected to the hybridization module 30 and to the estimation module 32, as shown in the figure 2 .

As an optional complement, the electronic processing device 20 further comprises a voice activity detection module 36 connected to the hybridization module 30.

In the example of the figure 1 , the electronic processing device 20 comprises an information processing unit 40 formed for example by a memory 42 and a processor 44 associated with the memory 42.

In the example of the figure 1 , the hybridization module 30, the estimation module 32, the noise reduction module 34, as well as in optional addition the voice activity detection module 36, are each produced in the form of software, or of a software brick, executable by the processor 44. The memory 42 of the processing device 20 is then capable of storing software for hybridizing the first and second analog signals into a hybrid signal, software for estimating the noise in the hybrid signal, and noise reduction software in the hybrid signal, as well as in optional addition voice activity detection software in the hybrid signal. The processor 44 is then capable of executing each of the software programs among the hybridization software, the estimation software and the noise reduction software, as well as in optional addition the voice activity detection software.

As a variant not shown, the hybridization module 30, the estimation module 32, the noise reduction module 34, as well as in optional addition the voice activity detection module 36, are each produced in the form of a programmable logic component, such as an FPGA ( Field Programmable Gate Array ) , or an integrated circuit, such as an ASIC ( Application Specifies Integrated Circuit ) .

When the electronic processing device 20 is produced in the form of one or more software programs, that is to say in the form of a computer program, also called a computer program product, it is also capable of being recorded on a medium, not shown, readable by computer. The computer-readable medium is for example a medium capable of storing electronic instructions and of being coupled to a bus of a computer system. For example, the readable medium is an optical disk, a magneto-optical disk, a ROM memory, a RAM memory, any type of non-volatile memory (for example EPROM, EEPROM, FLASH, NVRAM), a magnetic card or an optical card. A computer program comprising software instructions is then stored on the readable medium.

The hybridization module 30 is configured to calculate the hybrid signal from the first and second analog signals.

The hybridization module 30 is for example configured to obtain a first filtered signal by applying to the first signal a first filter associated with a first frequency range; to obtain a second filtered signal by applying to the second signal a second filter associated with a second frequency range; then to calculate the signal hybrid by summing the first filtered signal and the second filtered signal, the second frequency range being distinct from the first frequency range.

The first frequency range typically includes frequencies higher than those of the second frequency range; the first and second frequency ranges being for example disjoint.

The first filter is typically a high-pass filter with a cutoff frequency f _c substantially equal to 1000 Hz, the high-pass filter being for example a Gaussian high-pass filter. The second filter is typically a low-pass filter with a cutoff frequency also substantially equal to 1000 Hz, the low-pass filter being for example a Gaussian low-pass filter. In other words, the first frequency range is then the range of frequencies above 1000 Hz, and the second frequency range is that of frequencies below 1000 Hz.

In addition, the hybridization module 30 is configured to convert the first analog signal into a first digital signal, as the first analog signal is received, and to generate first successive sections from the first digital signal.

According to this addition, the hybridization module 30 is also configured to convert the second analog signal into a second digital signal, as the second analog signal is received, and to generate second successive sections from the second signal. digital.

According to this optional addition, the hybridization module 30 is then configured to calculate hybrid sections of the hybrid signal gradually, from the first and second sections generated; the corrected signal then being calculated from said hybrid sections.

In the example of the figure 2 , the hybridization module 30 comprises a first analog-digital converter 50, connected to the first aerial microphone 12 and configured to convert the first analog signal coming from the first microphone 12 into a first digital signal x _k ^aer , with a sampling frequency f _e for example substantially equal to 22 kHz. In addition, the first analog-digital converter 50 is configured to divide the first digital signal x _k ^aer , converted and sampled, into first successive sections, each first section comprising for example a number N of samples. The number N of samples in each first section is for example substantially equal to 512. Those skilled in the art will then observe that with the sampling frequency f _e substantially equal to 22 kHz and the number N of samples substantially equal to 512, the duration of each first section is approximately 20 ms, and typically substantially equal to 23 ms.

du premier signal numérique x_k ^aer, typiquement via une transformée de Fourier, telle qu'une transformée de Fourier rapide, également notée FFT (de l'anglais Fast Fourier Transform). Le module hybridation 30 comporte ensuite une première unité de filtrage 54, connectée en sortie du premier convertisseur temporel-fréquentiel 52 et configurée pour appliquer le premier filtre, typiquement le filtre passe-haut gaussien de fréquence de coupure f_c sensiblement égale à 1000 Hz, pour obtenir le premier signal filtré ${\tilde{X}}_k^{{\mathit{aer}}_{\mathit{HF}}}$

.In the example of the figure 2 , the hybridization module 30 further comprises a first time-frequency converter 52, connected to the output of the first analog-digital converter 50 and configured to calculate a first spectrum ${\tilde{X}}_k^{\mathit{air}}$

of the first digital signal x _k ^aer , typically via a Fourier transform, such as a fast Fourier transform, also denoted FFT ( Fast Fourier Transform ) . The hybridization module 30 then comprises a first filtering unit 54, connected to the output of the first time-frequency converter 52 and configured to apply the first filter, typically the Gaussian high-pass filter with a cutoff frequency f _c substantially equal to 1000 Hz, to obtain the first filtered signal ${\tilde{X}}_k^{{\mathit{air}}_{\mathit{HF}}}$

.

In the example of the figure 2 , the hybridization module 30 comprises a second analog-digital converter 60, connected to the second osteophonic microphone 14 and configured to convert the second analog signal coming from the second microphone 14 into a second digital signal x _k ^ost , with the sampling frequency f _e . In addition, the second analog-digital converter 60 is configured to divide the second digital signal x _k ^ost , converted and sampled, into second successive sections, each second section comprising for example the number N of samples. Those skilled in the art will then observe that with the sampling frequency f _e substantially equal to 22 kHz and the number N of samples substantially equal to 512, the duration of each second section is approximately 20 ms, and typically substantially equal to 23 ms.

du deuxième signal numérique x_k ^ost, typiquement via une transformée de Fourier, telle que la transformée de Fourier rapide, ou FFT. Le module hybridation 30 comporte ensuite une deuxième unité de filtrage 64, connectée en sortie du deuxième convertisseur temporel-fréquentiel 62 et configurée pour appliquer le deuxième filtre, typiquement le filtre passe-bas gaussien de fréquence de coupure f_c sensiblement égale à 1000 Hz, pour obtenir le deuxième signal filtré ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{BF}}}$

.In the example of the figure 2 , the hybridization module 30 further comprises a second time-frequency converter 62, connected to the output of the second analog-digital converter 60 and configured to calculate a second spectrum ${\tilde{X}}_k^{\mathit{ost}}$

of the second digital signal x _k ^ost , typically via a Fourier transform, such as the fast Fourier transform, or FFT. The hybridization module 30 then comprises a second filtering unit 64, connected to the output of the second time-frequency converter 62 and configured to apply the second filter, typically the Gaussian low-pass filter with a cutoff frequency f _c substantially equal to 1000 Hz, to obtain the second filtered signal ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{B.F.}}}$

.

By convention, in the present description, for a signal denoted x, its continuous form over time is denoted x(t), and its discretized form is denoted x[n] where n is a natural integer, n then forming a variable representing discretized time. In the frequency domain, m represents the discrete frequency variable, between 0 and N/2, where N represents the number of samples per section, for example equal to 512.

• où n est la variable entière représentant le temps discrétisé, et
• T_e est un pas de discrétisation temporelle vérifiant l'équation suivante : $$T_e=\frac{1}{f_e}$$
  
  où f_e est la fréquence d'échantillonnage, par exemple sensiblement égale à 22 kHz.

The discretized form of each signal then satisfies the following equation: $$x\left[\mathrm{not}\right]=x\left(\mathrm{not}\times T_e\right)$$

• where n is the integer variable representing the discretized time, and
• T _e is a temporal discretization step verifying the following equation: $$T_e=\frac{1}{f_e}$$
  
  where f _e is the sampling frequency, for example substantially equal to 22 kHz.

• où N est le nombre d'échantillons compris dans un tronçon,
• m est la variable de fréquence discrète, et
• f_e est la fréquence d'échantillonnage.

The discrete frequency variable m is typically associated with a frequency vector f[m] verifying the following equation: $$f\left[m\right]=m\times \frac{f_e}{\mathrm{NOT}}$$

• where N is the number of samples included in a section,
• m is the discrete frequency variable, and
• f _e is the sampling frequency.

The frequency then typically varies between 0 Hz and f _e /2 Hz, with a frequency step equal to f _e /N.

dans le domaine fréquentiel avec : $$\widetilde{X_k}\left[m\right]=\mathit{FFT}\left(x_k\left[n\right]\right)$$

où FFT représente l'opérateur numérique permettant d'estimer la transformée de Fourier discrète d'un signal, par exemple mis en oeuvre via le convertisseur temporel-fréquentiel 52, 62 respectif.By convention, the ^kth section of signal x is denoted x _k or x _k [n], and $\widetilde{X_k}\left[m\right]$

in the frequency domain with: $$\widetilde{X_k}\left[m\right]=\mathit{FFT}\left(x_k\left[\mathrm{not}\right]\right)$$

where FFT represents the digital operator making it possible to estimate the discrete Fourier transform of a signal, for example implemented via the respective time-frequency converter 52, 62.

représentant le spectre en amplitude et $\varphi \left(\widetilde{X_k}\left[m\right]\right)$

représentant le spectre en phase de x_k [n] respectivement. Par convention, le spectre sans autre précision désignera alors par la suite le spectre en amplitude.The spectral subtraction described below only requires working on the amplitude spectrum of the signal, the phase being preserved and unchanged throughout the process, with $\left|\widetilde{X_k}\left[m\right]\right|$

representing the spectrum in amplitude and $\varphi \left(\widetilde{X_k}\left[m\right]\right)$

representing the phase spectrum of x _k [ n ] respectively. By convention, the spectrum without further precision will then subsequently designate the amplitude spectrum.

et le deuxième signal filtré ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{BF}}}$

afin d'obtenir le signal hybride ${\tilde{X}}_k^{\mathit{hyb}}$

.In the example of the figure 2 , the hybridization module 30 also includes an adder 70, also called an adder, connected at the output of the first filtering unit 54, and the second filtering unit 64, and configured to sum the first signal filtered ${\tilde{X}}_k^{{\mathit{air}}_{\mathit{HF}}}$

and the second filtered signal ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{B.F.}}}$

in order to obtain the hybrid signal ${\tilde{X}}_k^{\mathit{hyb}}$

.

en sommant le premier signal filtré ${\tilde{X}}_k^{{\mathit{aer}}_{\mathit{HF}}}$

et le deuxième signal filtré ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{BF}}}$

via l'équation suivante : $${\tilde{X}}_k^{\mathit{hyb}}=\alpha {\tilde{X}}_k^{{\mathit{aer}}_{\mathit{HF}}}+\beta {\tilde{X}}_k^{{\mathit{ost}}_{\mathit{BF}}}$$

où α et β sont des constantes.The hybridization module 30 is then, for example, configured to calculate the hybrid signal ${\tilde{X}}_k^{\mathit{hyb}}$

by summing the first filtered signal ${\tilde{X}}_k^{{\mathit{air}}_{\mathit{HF}}}$

and the second filtered signal ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{B.F.}}}$

via the following equation: $${\tilde{X}}_k^{\mathit{hyb}}=\alpha {\tilde{X}}_k^{{\mathit{air}}_{\mathit{HF}}}+\beta {\tilde{X}}_k^{{\mathit{ost}}_{\mathit{B.F.}}}$$

where α and β are constants.

The values of the constants α and β are preferably adjustable making it possible to have an output signal at the level equivalent to that of the input of the first overhead microphone 12. In addition, this makes it possible to give a possible preponderance to the air signal, or respectively to the osteophonic signal.

As an optional complement, the hybridization module 30 is configured, during the generation of the first successive sections, to generate each new first section with samples of a first previous section and new samples of the first digital signal.

According to this optional complement, the hybridization module 30 is configured in a similar manner, during the generation of the second successive sections, to generate each new second section with samples of a previous second section and new samples of the second digital signal.

There is then an overlap between the first successive sections thus generated, that is to say from a first section generated to the next; and similarly between the second successive sections thus generated, that is to say from a second section generated to the next.

An overlap rate then corresponds to a ratio, within each new first section, between the number of samples from the previous first section used and the total number of samples from the first section, that is to say from the new first segment generated; or respectively to the ratio, within each new second section, between the number of samples from the previous second section used and the total number of samples from the second section. The overlap rate is for example between 50% and 75%, that is to say between 0.5 and 0.75. In other words, within each new first section, between half and three-quarters of the last samples from the previous first section are used; and similarly within each new second section, between half and three-quarters of the last samples from the previous second section are used. This overlap between sections is illustrated in Figure 3 .

On the Figure 3 , the sections which would be obtained by a simple division (ie without overlap) of the signal coming from the first analog-digital converter 50, respectively from the second analog-digital converter 60, are denoted x _i , which it concerns the first or second sections, where i is an index taking the successive values k-2, k-1 and k in this example. These sections x _i which would be obtained by simple cutting and without overlapping are also called physical sections. The other sections, represented in Figure 3 and illustrating this overlap, are also called overlapped sections and denoted x' _i , with i equal to k-1 or k in this example.

In the example of the Figure 3 , those skilled in the art will observe that the overlap rate is substantially equal to 50%, and that the section x' _k-1 then comprises 50% of samples from the previous section, corresponding to the last half of the section x _{k- 2} in this example; and 50% new samples, corresponding to the first half of the section x _k-1 in this example.

On the Figure 3 , the sections obtained after noise reduction by the noise reduction module 34 are denoted y _i when they result from physical sections x _i , and respectively y' _i when they result from overlapped sections x' _i , with i equal to k-1 or k in this example.

• où N représente le nombre d'échantillons par tronçon, par exemple égal à 512,
• y_i représente un tronçon obtenu après réduction de bruit à partir d'un tronçon physique x_i, et
• y'_i représente un tronçon obtenu après réduction de bruit à partir d'un tronçon chevauché x'_i.

In the case of a 50% overlap, the output section y _k ^out then typically verifies the following equation: $$y_k^{\mathit{out}}=\frac{1}{2}{y}_{k-1}^{\prime }\left[\frac{\mathrm{NOT}}{2}:\mathrm{NOT}\right]+\frac{1}{2}{y}_{k-1}\left[0:\mathrm{NOT}\right]+\frac{1}{2}{y}_k^{\prime }\left[0:\frac{\mathrm{NOT}}{2}\right]$$

• where N represents the number of samples per section, for example equal to 512,
• y _i represents a section obtained after noise reduction from a physical section x _i , and
• y' _i represents a section obtained after noise reduction from an overlapped section x' _i .

The estimation module 32 is configured to estimate noise in the hybrid signal.

When optional, the voice activity detection module 36 is configured to determine a presence of voice or an absence of voice in each section of the hybrid signal, the estimation module 32 is then configured to estimate the noise in the hybrid signal depending on each section with a determined absence of voice.

In other words, when the voice activity detection module 36 determines the presence of voices in a given section, the noise spectrum is not updated. On the contrary, when the voice activity detection module 36 determines the presence of voices in a given section, the background noise spectrum is updated. This update of the background noise spectrum is then carried out when the section is not voice and the The probability that this is noise is high. The robustness of the voice activity detection module 36 will make it possible to have even more precision in the estimation and tracking of noise.

si DAV = 0

• où p est un facteur d'oubli, de valeur par exemple égale à 0,95 ;
• DAV est un indicateur d'activité vocale issu du module de détection d'activité vocale 36, DAV étant égal à 1 si une présence de voix est déterminée, et à 0 sinon, i.e. si une absence de voix est déterminée ;
• $\left|{\tilde{X}}_k^{\mathit{hyb}}\right|$
  
  représente le spectre du signal hybride ${\tilde{X}}_k^{\mathit{hyb}}$
  
• $\left|{\tilde{N}}_{k-1}\right|$
  
  , et resp. |Ñ_k |, représentent les spectres du bruit de fond pour le tronçon d'indice k-1, et resp. d'indice k.

According to this optional complement, the estimation module 32 is typically configured to update the background noise spectrum | _Ñk | according to the following equation: $$\begin{array}{l}\{\left|\widetilde{{\mathrm{NOT}}_k}\right|=p\times \left|{\widetilde{\mathrm{NOT}}}_{k-1}\right|+\left(1-p\right)\times \left|{\tilde{X}}_k^{\mathit{hyb}}\right|\\ \left|\widetilde{{\mathrm{NOT}}_k}\right|=\left|{\widetilde{\mathrm{NOT}}}_{k-1}\right|\mathit{if}\mathit{DAV}=1\end{array}$$

if DAV = 0

• where p is a forgetting factor, with a value for example equal to 0.95;
• DAV is a voice activity indicator from the voice activity detection module 36, DAV being equal to 1 if a presence of voice is determined, and to 0 otherwise, ie if an absence of voice is determined;
• $\left|{\tilde{X}}_k^{\mathit{hyb}}\right|$
  
  represents the spectrum of the hybrid signal ${\tilde{X}}_k^{\mathit{hyb}}$
  
• $\left|{\widetilde{\mathrm{NOT}}}_{k-1}\right|$
  
  , and resp. | Ñ _k |, represent the background noise spectra for the section with index k-1, and resp. of index k.

The noise reduction module 34 is configured to calculate the corrected signal by applying a generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise.

In the example of the figure 2 , the noise reduction module 34 comprises a generalized spectral subtraction unit 80, also called SSG unit 80, capable of implementing the generalized spectral subtraction algorithm.

sinon

• |Ỹ_k [m] | représente le spectre du signal débruité pour le tronçon d'indice k ;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  représente le spectre du signal hybride pour le tronçon d'indice k ;
• $\left|\widetilde{N_k}\left[m\right]\right|$
  
  représente le spectre du bruit de fond pour le tronçon d'indice k ;
• α_k représente un coefficient de surestimation du bruit pour le tronçon d'indice k ;
• δ représente un coefficient de correction ;
• β représente un coefficient de réintroduction du bruit ; et
• γ représente un coefficient de puissance, typiquement égal à 1 ou 2.

The generalized spectral subtraction algorithm verifies for example the following equation: $$\begin{array}{l}\{{\left|\widetilde{Y_k}\left[m\right]\right|}^{\gamma }={\left|\widetilde{X_k}\left[m\right]\right|}^{\gamma }-{\alpha }_k\left[m\right]\times \delta \left[m\right]\times {\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|}^{\gamma }\mathrm{if}{\left|\widetilde{X_k}\left[m\right]\right|}^{\gamma }-{\alpha }_k\left[m\right]\times \delta \left[m\right]\times {\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|}^{\gamma }\ge \beta {\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|}^{\gamma }\\ {\left|\widetilde{Y_k}\left[m\right]\right|}^{\gamma }=\beta {\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|}^{\gamma }\end{array}$$

Otherwise

• | _Ỹk [m] | represents the spectrum of the denoised signal for the section of index k;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  represents the spectrum of the hybrid signal for the section of index k;
• $\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|$
  
  represents the background noise spectrum for the section of index k;
• α _k represents a noise overestimation coefficient for the section of index k;
• δ represents a correction coefficient;
• β represents a noise reintroduction coefficient; And
• γ represents a power coefficient, typically equal to 1 or 2.

The generalized spectral subtraction algorithm is calculated for example in amplitude, and the power coefficient γ is then equal to 1; or even in power, and the power coefficient γ is then equal to 2.

In the case of an amplitude calculation of the generalized spectral subtraction, with γ=1, little musical noise will be produced, but the estimated voice signal may be more or less distorted depending on the signal-to-noise ratio. Musical noise is a set of artifacts produced during spectral subtraction, consisting of short tones in time and producing a relatively unpleasant noise.

In the case of a power calculation of the generalized spectral subtraction, with γ =2, little distortion will be created, but a significant amount of musical noise may be generated.

The noise overestimation coefficient α is preferably recalculated at each section of index k, and is then denoted α _k . This coefficient helps prevent the generation of too much musical noise. To maximize its efficiency, its calculation is carried out by frequency bands and depends on the signal-to-noise ratio on each of these bands.

et $\left|\widetilde{N_k}\left[m\right]\right|$

sont d'abord découpés en sous-spectres notés $\left|\widetilde{X_k^J}\left[m\right]\right|$

et $\left|\widetilde{N_k^J}\left[m\right]\right|$

, où j représente le numéro de la bande de fréquence. Ainsi, j valeurs du rapport signal sur bruit, notées RSB_k ^j, chacune associée à une bande de fréquence d'indice j, sont typiquement calculées selon l'équation suivante : $${\mathit{RSB}}_k^j=10\times {\log }_{10}\left(\frac{{\displaystyle {\sum }_{m=0}^{N_j}{\left|\widetilde{X_k^J}\left[m\right]\right|}^2}}{{\displaystyle {\sum }_{m=0}^{N_j}{\left|\widetilde{N_k^J}\left[m\right]\right|}^2}}\right)$$

• où RSB_k ^j représente le rapport signal sur bruit pour le tronçon d'indice k et la bande de fréquence d'indice j,
• Nj représente le nombre d'échantillons fréquentiels contenus dans la bande d'indice j ;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  représente le spectre du signal hybride pour le tronçon d'indice k ; et
• $\left|\widetilde{N_k}\left[m\right]\right|$
  
  représente le spectre du bruit de fond pour le tronçon d'indice k.

The spectra $\left|\widetilde{X_k}\left[m\right]\right|$

And $\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|$

are first divided into subspectra noted $\left|\widetilde{X_k^J}\left[m\right]\right|$

And $\left|\widetilde{{\mathrm{NOT}}_k^J}\left[m\right]\right|$

, where j represents the frequency band number. Thus, j values of the signal-to-noise ratio, denoted SNR _k ^j , each associated with a frequency band of index j, are typically calculated according to the following equation: $${\mathit{RSB}}_k^j=10\times {\mathrm{<figure-callout\; id="10"\; label="log"\; filenames="imgf0001.png,imgf0006.png"\; state="\{\{state\}\}">log</figure-callout>}}_{10}\left(\frac{{\displaystyle {\sum }_{m=0}^{{\mathrm{NOT}}_j}{\left|\tilde{X_k^{\mathrm{<figure-callout\; id="2"\; label="J\; ̃\; m"\; filenames="imgf0003.png,imgf0005.png"\; state="\{\{state\}\}">J</figure-callout>}}}\left[m\right]\left[\right]\right|}^2}}{{\displaystyle {\sum }_{m=0}^{{\mathrm{NOT}}_j}{\left|\tilde{{\mathrm{NOT}}_k^{\mathrm{<figure-callout\; id="2"\; label="J\; ̃\; m"\; filenames="imgf0003.png,imgf0005.png"\; state="\{\{state\}\}">J</figure-callout>}}}\left[m\right]\left[\right]\right|}^2}}\right)$$

• where RSB _k ^j represents the signal-to-noise ratio for the section of index k and the frequency band of index j,
• Nj represents the number of frequency samples contained in the band of index j;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  represents the spectrum of the hybrid signal for the section of index k; And
• $\left|\widetilde{{\mathrm{NOT}}_k}\left[m\right]\right|$
  
  represents the background noise spectrum for the section with index k.

Then, for each signal-to-noise ratio value, the noise overestimation coefficient α _k verifies, for example, the following equation: $$\begin{array}{c}\{{\alpha }_k^j=4.75\mathit{if}{\mathit{RSB}}^j<-5\mathit{dB}\\ {\alpha }_k^j=4-\frac{3}{20}\times {\mathit{RSB}}_k^j\mathit{if}-5\mathit{dB}\le {\mathit{RSB}}_k^j\le 20\mathit{dB}\\ {\alpha }_k^j=1\mathit{if}{\mathit{RSB}}_k^j>20\mathit{dB}\end{array}$$

Overall, this calculation of the noise overestimation coefficient α makes it possible to overestimate the noise when the signal-to-noise ratio is low, and to reduce the introduction of musical noise type artifacts.

où l'intervalle [f_j ; f _j+1] correspond à toutes les fréquences de la j^ème bande de fréquences. Typiquement, à chaque tronçon la fonction α_k[m] sera une fonction constante par morceaux, où chaque morceau correspondra à une bande de fréquences déterminée par l'utilisateur.The noise overestimation coefficient α _k ^j is then converted so that it can be reintroduced into equation (8), for example according to the following equation: $${\alpha }_k\left[m\right]={\alpha }_k^j\forall m\in \left[\left\{f_j,f_{j+1}\right\}\right]$$

where the interval [ f _j ; f _{j +1} ] corresponds to all frequencies of the j ^th frequency band. Typically, at each chunk the function α _k [m] will be a piecewise constant function, where each chunk will correspond to a frequency band determined by the user.

The correction coefficient δ is a frequency correction coefficient calculated only once, typically at the start of the algorithm, and does not change over time.

This coefficient is a simple pre-factor depending on the frequency, in order to maximize certain frequency bands in a manner adapted to voice capture.

The correction coefficient δ is for example a piecewise constant function, verifying the following equation: $$\begin{array}{l}\{\delta \left[m\right]=1\forall f\left[m\right]<1000\mathit{Hz}\\ \delta \left[m\right]=2.5\forall f\left[m\right]\in [\mathrm{1000,2000}[\mathit{Hz}\\ \delta \left[m\right]=1.5\forall f\left[m\right]\in [\mathrm{2000,4000}[\mathit{Hz}\\ \delta \left[m\right]=1\forall f\left[m\right]\ge 4000\mathit{Hz}\end{array}$$

Taking into account the calculations carried out with the amplitude spectra, it is not necessary that the estimation | _Ỹk [ m ]| ^γ is negative because it would not make sense mathematically. This is why equation (8) includes a condition to avoid negative values.

The noise reintroduction coefficient β then makes it possible to choose whether or not to reintroduce noise in the event of potentially negative values. When the noise reintroduction coefficient β is chosen equal to 0, any subtraction leading to a negative value is replaced by the zero value. On the other hand, for any value greater than 0, noise is reintroduced. This keeps some of the noise that may be perceived as comfort noise masking some of the musical noise when any is created.

The noise reintroduction coefficient β is generally worth a few percent. The noise reintroduction coefficient β is for example substantially equal to 0.05, i.e. a reintroduction of 5% of the background noise into the output signal. This value is a predefined parameter.

It should be noted that the weaker or worse the signal-to-noise ratio, the less effective the estimation of the denoised signal is and the more the voice will be altered. It is therefore interesting to set a higher value for the noise reintroduction coefficient β in the case of a poor signal-to-noise ratio, in order to recapture some harmonics of the voice in the background noise which would otherwise be lost in the spectral subtraction. .

In the example of the figure 2 , the noise reduction module 34 further comprises a frequency-time converter 82, connected to the output of the generalized spectral subtraction unit 80, and configured to calculate a time signal from the frequency signal coming from the SSG unit 80 , typically via an inverse Fourier transform, such as an inverse fast Fourier transform, also denoted IFFT ( Inverse Fast Fourier Transform ) .

• où y_k[n] représente le signal de sortie débruité pour le tronçon d'indice k ;
• IFFT représente l'opérateur numérique de transformée de Fourier inverse ;
• |Ỹ_k [m]| , et resp. $\varphi \left(\widetilde{X_k}\left[m\right]\right)$
  
  , représentent le spectre en amplitude, et resp. en phase, du signal débruité pour le tronçon d'indice k.

As stated previously, the frequency domain calculations were carried out with the amplitude of the signal spectrum of the section. The phase of the latter, which remains unmodified, is then reintegrated into the signal before the inverse Fourier transform making it possible to return to the time domain, for example according to the following equation: $$y_k\left[\mathrm{not}\right]=\mathit{IFFT}\left(\left|\widetilde{Y_k}\left[m\right]\right|\times e^{\mathit{j\varphi }\left(\widetilde{X_k}\left[m\right]\right)}\right)$$

• where y _k [n] represents the denoised output signal for the section of index k;
• IFFT represents the inverse Fourier transform digital operator;
• | _Ỹk [ m ]| , and resp. $\varphi \left(\widetilde{X_k}\left[m\right]\right)$
  
  , represent the amplitude spectrum, and resp. in phase, of the denoised signal for the section of index k.

In the example of the figure 2 , the noise reduction module 34 then comprises a digital-analog converter 84, connected to the output of the frequency-time converter 82 and configured to provide the corrected signal y(t) in analog form. The denoised signal y _k ^hyb from the frequency-time converter 82 is then resynthesized into the corrected signal y(t) via the digital-analog converter 84, with synthesis of the overlapped sections if necessary, then delivered at the output of the processing device 20 .

The voice activity detection module 36 is configured to determine a presence of voice or an absence of voice in each section of the hybrid signal.

The voice activity detection module 36 is for example configured to determine the presence of voice or the absence of voice from the second signal from the bone mechanical excitation transducer; and preferably only from said second signal, without taking into account the first signal.

The second microphone 14, osteophonic or solid, is capable of measuring the vibrations of the skin and the face linked to the solicitation of the vocal cords, and makes it possible to capture the voiced part of a vocal signal while being very insensitive to the noise of background (which a priori does not vibrate the user's skin enough to be captured).

The advantage of using the second osteophonic microphone 14 lies in its immunity to background noise. This immunity is even greater in the low frequency part of the acquired signal.

issu de la deuxième unité de filtrage 64.Advantageously, the detection of voice activity is then carried out after filtering in the frequency domain (also operating in the time domain) of the solid-body signal. The voice activity detection module 36 is then preferably configured to determine the presence of voice or the absence of voice from the second filtered signal from the second filtered signal. ${\tilde{X}}_k^{{\mathit{ost}}_{\mathit{B.F.}}}$

from the second filtering unit 64.

As an optional complement, the voice activity detection module 36 is configured to calculate an RMS value for each section of the second signal, i.e. for each second section; then to determine the presence of voice or absence of voice based on respective RMS values.

The processing is based on the calculation of the signal energy section by section. However here, thanks to the noise-immune nature of the filtered solid-state microphone signal, the energy of the voice will emerge all the time from the noise floor energy. Calculating the RMS level then allows us to know the energy of the signal.

As known per se, the effective value, also called RMS value (from the English Root Mean Square, meaning square mean), of a periodic signal is the square root of the average of the square of this quantity, over a time interval given or the square root of the moment of order two (or variance) of the signal.

• où RMS_k représente la valeur RMS pour le tronçon d'indice k ;
• x_k [n] représente le signal pour le tronçon d'indice k ;
• N représente le nombre d'échantillons dudit tronçon.

For a time segment x _k [ n ] of N samples, the calculation of the RMS value is then typically carried out via the following equation: $${\mathit{RMS}}_k=\sqrt{\frac{{\displaystyle {\sum }_{\mathrm{not}=0}^{\mathrm{NOT}}{x}_k{\left[\mathrm{not}\right]}^2}}{\mathrm{NOT}}}$$

• where RMS _k represents the RMS value for the section of index k;
• x _k [ n ] represents the signal for the section of index k;
• N represents the number of samples of said section.

• où RMS_k représente la valeur RMS pour le tronçon d'indice k ;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  représente le spectre du signal hybride pour le tronçon d'indice k ; et
• N représente le nombre d'échantillons dudit tronçon.

However, in the frequency domain, thanks to the Parseval identity according to which the energy is equal in the frequency and time domains, we obtain the following equation: $${\mathit{RMS}}_k=\frac{1}{2\mathrm{NOT}}\sqrt{{\displaystyle \sum _{m=-\frac{\mathrm{NOT}}{2}}^{\frac{\mathrm{NOT}}{2}}{\left|\tilde{X_{\mathrm{<figure-callout\; id="2"\; label="k\; ̃\; m"\; filenames="imgf0003.png,imgf0005.png"\; state="\{\{state\}\}">k</figure-callout>}}}\left[m\right]\left[\right]\right|}^2}}$$

• where RMS _k represents the RMS value for the section of index k;
• $\left|\widetilde{X_k}\left[m\right]\right|$
  
  represents the spectrum of the hybrid signal for the section of index k; And
• N represents the number of samples of said section.

où log₁₀ représente l'opérateur logarithme décimal, ou encore logarithme de base 10.This RMS level value is optionally converted into a dBFS value from the following equation: $${\mathit{RMS}}_k^{\mathit{dB}}=20\times {\mathrm{<figure-callout\; id="10"\; label="log"\; filenames="imgf0001.png,imgf0006.png"\; state="\{\{state\}\}">log</figure-callout>}}_{10}\left({\mathit{RMS}}_k\right)$$

where log ₁₀ represents the decimal logarithm operator, or logarithm to base 10.

This dBFS value is typically between -94 dBFS at least (in the case of a dynamic resolution of 16 bits) and 0 dBFS at maximum (for a constant signal which would be worth 1).

As an optional addition, the voice activity detection module 36 is configured to determine the presence of voices or the absence of voices based on an average value of M last calculated RMS values, also called smoothed RMS, and/or of a variation in RMS value between a current RMS value and a previous RMS value, also called rate of variation of the RMS level, with M an integer greater than or equal to 1.

According to this optional addition again, the voice activity detection module 36 is for example configured to determine the presence of voices if said average value is greater than or equal to a predefined average threshold A or if said RMS value variation is greater than or equal to a predefined variation threshold B.

, vérifie par exemple l'équation suivante : $$\overline{{\mathit{RMS}}_k^{\mathit{dB}}}=\frac{1}{M}\times {\displaystyle \sum _{j=0}^{M-1}{\mathit{RMS}}_{k-j}^{\mathit{dB}}}$$

The value of the RMS level is likely to vary over time, and to undergo sudden variations when the microphone concerned, in particular the second microphone 14, picks up a significant vibration. This optional addition then makes it possible to improve the precision and reduce the errors of the algorithm, with averaging over the last M calculated values of the RMS level (during the last M sections). This is for example implemented via a circular buffer which for each new section adds the new calculated RMS value, deletes the last ^{Mth value} , then averages the old one. The smoothed RMS level at the ^kth section, noted $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$

, verifies for example the following equation: $$\overline{{\mathit{RMS}}_k^{\mathit{dB}}}=\frac{1}{M}\times {\displaystyle \sum _{j=0}^{M-1}{\mathit{RMS}}_{k-j}^{\mathit{dB}}}$$

au cours du temps permet de repérer les zones de voix lorsque celui-ci dépasse un certain seuil. Néanmoins, dû au lissage, ce niveau peut dépasser le seuil choisi légèrement en retard. Avantageusement, une deuxième métrique liée au niveau au RMS, à savoir le taux de variation du niveau RMS noté ΔRMS_k ^dB, est alors calculée pour mieux détecter l'apparition de la voix, par exemple via l'équation suivante : $${\mathit{\Delta RMS}}_k^{\mathit{dB}}=\frac{\left(\overline{{\mathit{RMS}}_k^{\mathit{dB}}}-\overline{{\mathit{RMS}}_{k-1}^{\mathit{dB}}}\right)}{\mathit{dt}}$$

• où ΔRMS_k ^dB représente le taux de variation du niveau RMS pour le tronçon d'indice k ;
• $\overline{{\mathit{RMS}}_{k-1}^{\mathit{dB}}}$
  
  , resp. $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$
  
  , représente le niveau RMS lissé pour le tronçon d'indice k-1, et resp. d'indice k ;
• dt représente un delta de temps entre deux tronçons successifs.

Monitoring the value of $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$

over time makes it possible to identify areas of voice when it exceeds a certain threshold. However, due to smoothing, this level may exceed the chosen threshold slightly late. Advantageously, a second metric linked to the RMS level, namely the rate of variation of the RMS level denoted ΔRMS _k ^dB , is then calculated to better detect the appearance of the voice, for example via the following equation: $${\mathit{\Delta RMS}}_k^{\mathit{dB}}=\frac{\left(\overline{{\mathit{RMS}}_k^{\mathit{dB}}}-\overline{{\mathit{RMS}}_{k-1}^{\mathit{dB}}}\right)}{\mathit{dt}}$$

• where ΔRMS _k ^dB represents the rate of variation of the RMS level for the section of index k;
• $\overline{{\mathit{RMS}}_{k-1}^{\mathit{dB}}}$
  
  , resp. $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$
  
  , represents the smoothed RMS level for the section with index k-1, and resp. of index k;
• dt represents a time delta between two successive sections.

The value dt can correspond exactly to the time delta between two successive sections, and the variation of the RMS level will then be expressed in dB.s ^-1 , but this can take very large values.

Alternatively, and for convenience, the value dt is chosen equal to 1. Where appropriate, ΔRMS _k ^dB is a rate of variation expressed in dB.section ^-1 . This quantity is relevant because the moment an interlocutor begins to speak, the RMS level increases suddenly, resulting in a positive ΔRMS _k ^dB greater than 1 dB.section ^-1 . This size varies quickly, it makes it possible to detect the voice very quickly, thus avoiding missing the start of a sentence.

• où $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$
  
  représente le niveau RMS lissé pour le tronçon d'indice k ;
• ΔRMS_k ^dB représente taux de variation du niveau RMS pour le tronçon d'indice k ;
• DAV_k est un indicateur d'activité vocale pour le tronçon d'indice k, cet indicateur étant égal à 1 si une présence de voix est déterminée, et à 0 sinon ;
• A représente le seuil prédéfini de moyenne et B représente le seuil prédéfini de variation, correspondant respectivement aux seuils de niveau et du taux de variation à dépasser pour considérer que le tronçon est parlé.

Decision making for instantaneous voice activity detection is then defined for example by the following equation: $$\begin{array}{c}\{\mathrm{If}\overline{{\mathit{RMS}}_k^{\mathit{dB}}}\ge \mathrm{HAS}\mathrm{SO}{\mathit{DAV}}_k=1\\ \mathrm{Or}\mathrm{if}{\mathit{\Delta RMS}}_k^{\mathit{dB}}\ge B\mathrm{SO}{\mathit{DAV}}_k=1\\ \mathrm{Otherwise}{\mathit{DAV}}_k=0\end{array}$$

• Or $\overline{{\mathit{RMS}}_k^{\mathit{dB}}}$
  
  represents the smoothed RMS level for the section of index k;
• ΔRMS _k ^dB represents rate of variation of the RMS level for the section of index k;
• DAV _k is a voice activity indicator for the section of index k, this indicator being equal to 1 if the presence of voice is determined, and to 0 otherwise;
• A represents the predefined average threshold and B represents the predefined variation threshold, corresponding respectively to the level and rate of variation thresholds to be exceeded to consider that the section is spoken.

These threshold values A and B are predefined as a function of the dynamics of the acoustic device 10, for example as a function of the gain of the microphone concerned, in particular of the second microphone 14, etc.

The voice activity detection calculation described above gives an instantaneous value for each successive chunk (whether overlapped or not). Relying solely on an instantaneous value can lead to errors, for example a micro-silence in the voice could create an unwanted change to 0 in the DAV voice activity indicator. On the contrary, a very short impulse noise can lead to a DAV voice activity indicator equal to 1 for a single section, before returning to 0. Depending on the use of the voice activity detection module 36 (with a mode where the channel is only open if DAV = 1 for example), this behavior can cause unpleasant artifacts. This is why the calculation of voice activity detection is advantageously smoothed in order to avoid these artifacts.

This smoothing is for example carried out using an attack time and a release time. When an instantaneous DAV voice activity indicator DAV _inst ^k is equal to 1 at least as long as the attack time (or the equivalent number of chunks), then a smoothed DAV voice activity indicator _smooth DAV ^k becomes equal to 1. On the contrary, when the voice activity indicator DAV instantaneous DAV _inst ^k is equal to 0 at least as long as the release time, then the voice activity indicator DAV smoothed DAV _smooth ^k returns to 0. In all other cases, the smoothed DAV voice activity indicator _smoothed DAV ^k retains the value it had at the previous section. To implement this smoothing, a counter C _k is for example used. The modification of this counter C _k is typically governed by Table 1 below for each current section of index k, as a function of the instantaneous DAV voice activity indicator DAV _inst ^k and the value of the counter C _{k- 1} to the previous section of index k-1: [Table 1] AND C _k-1 ≥ 0 C _k-1 < 0 DAV _ins t ^k = 0 Counter reset: C _k = 0 C _k = C _k-1 -1 DAV _ins t ^k = 1 C _k = C _k-1 +1 Counter reset: C _k = 0

• où DAV_lisse ^k est l'indicateur d'activité vocale lissé pour le tronçon d'indice k, cet indicateur étant égal à 1 si une présence de voix est déterminée, et à 0 sinon ;
• C_k est le compteur pour le tronçon d'indice k ;
• t_atk représente le temps d'attaque ; et
• t_rel représente le temps de relâche.

The decision-making for the detection of smoothed vocal activity is then defined for example by the following equation: $$\begin{array}{c}\{\mathrm{If}{\mathrm{VS}}_k>t_{\mathit{atk}}\mathrm{SO}{\mathit{DAV}}_{\mathit{smooth}}^k=1\\ \mathrm{If}{\mathrm{VS}}_k<-t_{\mathit{rel}}\mathrm{SO}{\mathit{DAV}}_{\mathit{smooth}}^k=0\\ \mathrm{Otherwise}{\mathit{DAV}}_{\mathit{smooth}}^k={\mathit{DAV}}_{\mathit{smooth}}^{k-1}\end{array}$$

• where _smooth DAV ^k is the smoothed voice activity indicator for the section of index k, this indicator being equal to 1 if a presence of voice is determined, and to 0 otherwise;
• C _k is the counter for the section with index k;
• t _atk represents the attack time; And
• t _rel represents the release time.

The operation of the acoustic device 10, and in particular of the processing device 20, according to the invention will now be described with regard to the figure 4 representing a flowchart of the treatment method according to the invention.

The processing applied to the signal to reduce noise is carried out digitally and in real time. Indeed, when the operator uses the acoustic device 10, the signal must be denoised and sent to the interlocutor as quickly as possible, seeking to reduce the latency as much as possible, with a desired value of 20 to 30 ms. To enable qualitative denoising, there must be a minimum amount of information to analyze before the noise can be reduced effectively. The processing carried out is then block processing, applied section by section to the input signal. As indicated previously, the chunks are typically each approximately 20 ms long. Indeed, over this duration, the voice behaves almost stationary, whereas noise does so over much longer durations.

In order to optimize power consumption, the sampling frequency is preferably less than 22,050 Hz, allowing a bandwidth in the interval [0; 11,025 Hz]. Consequently, to have signal sections of approximately 20 ms at this sampling frequency, these must typically contain 512 samples.

The processing applied to the signal to reduce noise is largely carried out in the frequency domain, which is more suitable for denoising because the aim is to reduce the level in the frequency bands containing the most noise. However, due to working in frequency sections, problems of discontinuities and inaccuracies can appear from one section to another, and an overlap of sections, with an overlap rate preferably greater than 50%, ideally equal to 75%, as described above, is then advantageously used to attenuate them.

During an initial step 100, the processing device 20 then calculates, via its hybridization module 30, the hybrid signal from the first and second analog signals, coming from the first and second microphones 12, 14, in the manner described previously.

During a following optional step 110, the processing device 20 determines, via its voice activity detection module 36, a presence of voice or an absence of voice in each section of the hybrid signal, in the manner described above.

The processing device 20 then estimates, during the next step 120 and via its estimation module 32, the noise in the hybrid signal, obtained previously during the hybridization step 100, in the manner described above.

When optionally a presence of voice or an absence of voice in each section of the hybrid signal has been determined during the voice activity detection step 110, the noise is then, during the estimation step 120, estimated in the hybrid signal according to each section with a determined absence of voice, in the manner described previously.

Finally, during the next step 130, the processing device 20 applies, via its noise reduction module 34, the generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise, in order to calculate the corrected signal.

As indicated previously, the processing method is in real time or near real time, with a latency of approximately 20 to 30 ms, and block processing, applied section by section to the input signal.

Also, at the end of step 130, the processing process returns to the initial step 100, and more generally, each of the steps 100, optionally 110, 120 and 130 is repeated regularly in order to be implemented for each successive signal section.

On the figure 5 , curve 200 then represents an example with a signal coming from an aerial recording of a speaker speaking in a highly noisy environment (vehicle noise at more than 90 db(A)). The 250 curve at the figure 5 presents the same signal after the implementation of the processing device 20 according to the invention. We note that the noise is greatly attenuated with the processing device 20 according to the invention, while observing that the parts corresponding to the voice are clearly visible and then have good intelligibility.

There Figure 6 presents an example of voice activity detection used on a voice signal recorded by a conventional overhead microphone for different successive phases of noise, from no noise to loud noise. The curve 300 is the temporal representation of this signal on which the decision taken by the detection of voice activity is superimposed, where the gray areas 310 correspond to areas for which the presence of voice has been determined, ie DAV = 1; the other zones, not grayed or white, corresponding to zones for which an absence of voice has been determined, ie DAV = 0. On the Figure 6 , the curve 320 represents the RMS level of this signal coming from the aerial microphone over time with the threshold level to be exceeded for taking decision, the threshold level being represented by the horizontal line 330 in dotted line. Curve 340 corresponds to the algorithm's estimation of the RMS level of the background noise in the phases where the detection of voice activity has determined an absence of voice.

In this example of the Figure 6 , the threshold level was deliberately chosen low, with a value approximately equal to -40 dBFS to allow good voice detection in the absence of noise. Indeed, we see that in the noise-free phase, for the time period between time instants 0s and 15s, the voice emerges from the noise and the averaged RMS level exceeds the threshold each time the user speaks. Classic voice activity detection is therefore correct on the silent part. However, as soon as the noise presents a moderate level, the averaged RMS level is systematically above the set threshold, since it is too low. Consequently, this results in an erroneous determination of the presence of voices throughout the rest of the signal: the detection of vocal activity then becomes ineffective, because it is incapable of separating the contribution of the noise from that of the voice. Since the detection of voice activity always gives a positive response, the estimate of the RMS level of the noise is also completely distorted, and remains at the value taken when there was no noise.

There Figure 7 is analogous to the Figure 6 , with the difference that the detection threshold has been raised to a value approximately equal to -20 dBFS. The curve 400 is the temporal representation of this signal on which the decision taken by the detection of voice activity is superimposed, where the gray areas 410 correspond to areas for which the presence of voice has been determined, ie DAV = 1; the other zones, not grayed or white, corresponding to zones for which an absence of voice has been determined, ie DAV = 0. On the Figure 7 , the curve 420 represents the RMS level of this signal coming from the aerial microphone over time with the threshold level to be exceeded for decision-making, the threshold level being represented by the horizontal line 430 in dotted line. Curve 440 corresponds to the algorithm's estimation of the RMS level of the background noise in the phases where the detection of voice activity has determined an absence of voice.

On the Figure 7 , those skilled in the art will then note that voice detection in the part with moderate noise, between approximately time points 15s and 30s, is rather correct. The RMS level, at times when there is voice, allows it to be discriminated from noise. However, as soon as the noise level is further increased, this threshold no longer makes it possible to clearly distinguish the voice from the noise, and many areas are considered as exclusively spoken, between the temporal instants 34s and 42s for example, whereas there are actually moments of voicelessness in these areas. Worse still, due to the threshold being too high, in the noise-free part, state-of-the-art voice activity detection confuses the voice with noise several times and misses some detections or cuts them off. too early. This then seriously deteriorates the voice signal. In addition, this completely distorts the estimate of the noise level, corresponding to curve 440, which is artificially increased when the person speaks.

Finally, through these two examples of figures 6 And 7 illustrating the state of the art, those skilled in the art will understand that the threshold should vary automatically (low for the silent phases, higher for the noise phases) to allow good activity detection results state of the art voice with an overhead microphone. Indeed, with traditional voice activity detection, a fixed threshold setting cannot correctly correspond to both a noisy environment and a quiet environment, in particular due to the high sensitivity of aerial microphones to the environment.

There figure 8 illustrates the implementation of the processing device 20 according to the invention, and in particular the detection of vocal activity according to the invention from the second signal coming from the transducer with mechanical bone excitation, this on the same recording as that used for the examples of figures 6 And 7 , but with the second osteophonic microphone 14, and then the use of the generalized spectral subtraction algorithm.

The curve 500 is the temporal representation of this signal on which the decision taken by the detection of voice activity is superimposed, where the gray areas 510 correspond to areas for which the presence of voice has been determined, ie DAV = 1; the other zones, not grayed or white, corresponding to zones for which an absence of voice has been determined, ie DAV = 0. On the figure 8 , the curve 520 represents the RMS level of this signal coming from the second osteophonic microphone 14 over time with the threshold level to be exceeded for decision-making, the threshold level being represented by the horizontal line 530 in dotted line. Curve 540 corresponds to the algorithm's estimation of the RMS level of the background noise in the phases where the detection of voice activity has determined an absence of voice.

With the processing device 20 according to the invention, a first striking element is that the waveform associated with this filtered osteophonic recording (low-pass filter) is much less marked by noise. Regardless of the noise level, the voice emerges very easily from it. This effect is even more visible on the representation of the RMS level of the filtered signal over time, there is almost 40 dB difference between the peaks linked to the voice and the background noise. Consequently, the choice of the threshold value becomes easier and offers greater latitude than with the processing device of the state of the art. This threshold was for example set arbitrarily here at -35dBFS, while observing that a threshold value at -25dBFS or -45dBFS would have given similar results. Thanks to this natural emergence, the generalized spectral subtraction algorithm is particularly effective and identifies the voice equally well in three different noise zones.

Finally, thanks to its performance, the processing device 20 according to the invention is capable of precisely detecting the temporal periods in the presence of noise only. In this way, averaging the RMS level of the overhead microphone only at times when DAV = 0 makes it possible to obtain a good estimate of the background noise level, represented by curve 540.

These results clearly show the interest of the processing device 20 according to the invention due to the significant gain in performance and calculation cost, compared to the processing device of the state of the art.

Thus, when the user is in a noisy environment, and he uses the acoustic device 10, for example with a radio, to communicate with a remote interlocutor, the signal sent to the interlocutor would be, without implementation work of the invention, altered by the unwanted capture of a portion of background noise. The electronic processing device 20 according to the invention makes it possible to reduce the presence of this background noise in the signal sent to the interlocutor, and in particular to filter the voice from this noise, in order to aim to send only the signal useful to the interlocutor via the radio.

The results obtained with the electronic processing device 20 according to the invention, in particular those presented above with regard to the figures 5 And 8 , further show the synergy between the detection of vocal activity based on the capture of a signal via the second osteophonic microphone 14 and the reduction of noise via the generalized spectral subtraction algorithm. This synergy allows for very good precision regarding vocal activity, which allows the noise spectrum to be updated effectively. The results obtained with the generalized spectral subtraction algorithm are then improved, while using a limited number of calculation operations.

We can thus see that the electronic processing device 20, and the associated processing method, make it possible to further improve the reduction of noise in the signal delivered at the output of the acoustic device 10.

Claims (10)

Electronic processing device (20) for an acoustic device (10), the acoustic apparatus (10) comprising a first microphone (12) comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from a user's vocal cords and of transforming said acoustic waves into a first analog signal; and a second microphone (14) comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal, the electronic processing device (20) being configured to be connected to the first and second microphones (12,14), to receive the first and second analog signals as input and to output a corrected signal, the electronic processing device (20) comprising: - a hybridization module (30) configured to calculate a hybrid signal from the first and second analog signals; characterized in that it further comprises: - an estimation module (32) connected to the hybridization module (30) and configured to estimate noise in the hybrid signal; - a noise reduction module (34) connected to the hybridization module (30) and to the estimation module (32), the noise reduction module (34) being configured to calculate the corrected signal by applying an algorithm generalized spectral subtraction of the hybrid signal and as a function of the estimated noise. Device (20) according to claim 1, in which the hybrid signal comprises several successive sections, and the device (20) further comprises a voice activity detection module (36) connected to the hybridization module (30) and configured to determine a presence of voice or an absence of voice in each section of the hybrid signal; the estimation module (32) then being configured to estimate the noise in the hybrid signal as a function of each section with a determined absence of voice. Device (20) according to claim 2, in which the voice activity detection module (36) is configured to determine the presence of voice or the absence of voice from the second signal from the bone mechanical excitation transducer;
the voice activity detection module (36) preferably being configured to determine the presence of voice or the absence of voice solely from the second signal, without taking into account the first signal. Device (20) according to claim 3, in which the second signal comprises several successive sections, and the voice activity detection module (36) is configured to calculate an RMS value for each section of the second signal, then to determine the presence voice or absence of voice depending on respective RMS value(s). Device (20) according to claim 4, wherein the voice activity detection module (36) is configured to determine the presence of voices or the absence of voices based on an average value of M last value(s) (s) calculated RMS and/or a variation in RMS value between a current RMS value and a previous RMS value, M being an integer greater than or equal to 1;
the voice activity detection module (36) preferably being configured to determine the presence of voices if said average value is greater than or equal to a predefined average threshold (A) or if said RMS value variation is greater than or equal to a predefined variation threshold (B). Device (20) according to any one of the preceding claims, in which the hybridization module (30) is configured to convert the first analog signal into a first digital signal, as the first analog signal is received, and to generate first successive sections from the first digital signal, each new first section generated comprising samples of a first previous section and new samples of the first digital signal; And the hybridization module (30) is configured to convert the second analog signal into a second digital signal, as the second analog signal is received, and to generate second successive sections from the second digital signal, each new second section generated comprising samples of a previous second section and new samples of the second digital signal; hybrid sections of the hybrid signal then being calculated progressively from the first and second sections generated; the corrected signal then being calculated from said hybrid sections. Device (20) according to any one of the preceding claims, in which the hybridization module (30) is configured to obtain a first filtered signal by applying to the first signal a first filter associated with a first frequency range; to obtain a second filtered signal by applying to the second signal a second filter associated with a second frequency range; then to calculate the hybrid signal by summing the first filtered signal and the second filtered signal, the second frequency range being distinct from the first frequency range; the first frequency range preferably comprising frequencies higher than those of the second frequency range; the first and second frequency ranges being preferably still disjoint. Acoustic device (10) comprising: - a first microphone (12) comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from the vocal cords of a user and of transforming said acoustic waves into a first analog signal; - a second microphone (14) comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal; - an electronic processing device (20) connected to the first and second microphones (12,14), the electronic processing device (20) being configured to receive the first and second analog signals as input, then to output a corrected signal ; characterized in that the electronic processing device (20) is according to any one of the preceding claims. Processing method, the method being implemented by an electronic processing device (20) connected to first and second microphones (12,14), the first microphone (12) comprising an electroacoustic transducer capable of receiving acoustic sound waves of a sound signal coming from a user's vocal cords and transforming said acoustic waves into a first analog signal; and the second microphone (14) comprising a transducer with mechanical bone excitation capable of receiving by bone conduction vibrational oscillations of said sound signal and of transforming said vibrational oscillations into a second analog signal, the electronic processing device (20) being configured to receive as input the first and second analog signals and to output a corrected signal, the treatment process comprising: - a hybridization step (100) comprising the calculation of a hybrid signal from the first and second analog signals; characterized in that it further comprises: - a step of estimating (120) a noise in the hybrid signal; And - a noise reduction step (130) comprising the calculation of the corrected signal by applying a generalized spectral subtraction algorithm to the hybrid signal and as a function of the estimated noise. Computer program comprising software instructions which, when executed by a computer, implement a method according to the preceding claim.

Images