EP0710947A1 - Method and apparatus for noise suppression in a speech signal and corresponding system with echo cancellation - Google Patents

Method and apparatus for noise suppression in a speech signal and corresponding system with echo cancellation Download PDF

Info

Publication number
EP0710947A1
EP0710947A1 EP95402385A EP95402385A EP0710947A1 EP 0710947 A1 EP0710947 A1 EP 0710947A1 EP 95402385 A EP95402385 A EP 95402385A EP 95402385 A EP95402385 A EP 95402385A EP 0710947 A1 EP0710947 A1 EP 0710947A1
Authority
EP
European Patent Office
Prior art keywords
signal
speech signal
noise
frequency
processing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
EP95402385A
Other languages
German (de)
French (fr)
Other versions
EP0710947B1 (en
Inventor
Ivan Bourmeyster
Frédéric Lejay
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alcatel Lucent SAS
Original Assignee
Alcatel Mobile Communication France SA
Alcatel Mobile Phones SA
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Alcatel Mobile Communication France SA, Alcatel Mobile Phones SA filed Critical Alcatel Mobile Communication France SA
Publication of EP0710947A1 publication Critical patent/EP0710947A1/en
Application granted granted Critical
Publication of EP0710947B1 publication Critical patent/EP0710947B1/en
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208—Noise filtering
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
    • Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10S367/00—Communications, electrical: acoustic wave systems and devices
    • Y10S367/901—Noise or unwanted signal reduction in nonseismic receiving system

Definitions

  • the present invention relates generally to method and device for suppressing a noise signal in a speech signal, typically for an application to hands-free radiotelephony. It also relates to a system implementing such a device in combination with an echo canceller.
  • the electrical signal resulting from an acousto-electric conversion of a signal speech is mixed with a noise signal.
  • the noise level in this environment is high, for example in the passenger compartment of a vehicle, it is necessary to implement treatments aimed at suppress the noise signal in the electrical signal of speech.
  • two types of removal processing are distinguished according to the prior art: the spectral subtraction processing and the so-called processing by filter bank.
  • Treatment by filter bank consists of a step of separating the input signal into a plurality of representative time signals each of a band of respective predetermined frequencies, an estimation step a signal-to-noise ratio for each of these signals temporal, a step of weighting these temporal signals by respective coefficients which are each a function of the respective one signal to noise ratios for the signal time considered, and a step of adding these signals time weighted as a resulting signal consisting of a speech signal in which the noise signal is deleted.
  • each of the signal-to-noise ratios is estimated based on the variation of the power of the time signal considered in its frequency band respective.
  • Such a treatment by filter bank requires considerable means of calculation due to the fact that all stages of separation, estimation, weighting and above addition are carried out in the time domain.
  • the means of calculation being limited, in terms of Millions of Instructions Per Second (Mips), by the capacity of the processing processor digital signal (DSP in English terminology for Digital Signal Processor), it is then proposed to reduce noise signal suppression processing at bands coarse frequencies, and therefore reduce the fineness, or precision, of this treatment.
  • Mips Millions of Instructions Per Second
  • DSP Digital Signal Processor
  • Spectral subtraction processing uses its part in the frequency domain, typically by use Fast Fourier Transform (FFT) in Anglo-Saxon terminology). It has the disadvantage main to induce nonlinear distortion in the processed speech signal resulting from loss phase information of this signal. Indeed this treatment by spectral subtraction produces such a distortion because it applies to samples resulting from the Transformation of Fast Fourier of the noisy speech signal to process squared module functions that suppress information from phase, making this treatment non-linear.
  • the lack of linearity of the subtraction processing spectral prevents its effective use in combination with echo cancellation processing, such as as proposed by the invention, because the cancellation device echo is disturbed in its functioning by this loss phase information.
  • a first objective of the present invention is to provide a method of suppressing noise in a signal speech which has the advantage of considerably reducing the computing power required in terms of Number Instructions Per Second, Compared To Treatment by filter bank.
  • a second objective of the invention is to provide a process which does not induce non-linear distortion in the speech signal to be processed, in contrast to the processing by spectral subtraction.
  • Another object of the invention is to provide a system including a noise canceling device implementing the process steps, in combination with an echo canceller.
  • the calculation step is preceded, for each of the K groups of frequency components, by a step of selecting a part of the components frequencies with respective predetermined ranks in said each of the groups, said selected part presenting a character of symmetry with respect to the complementary to this part among the plurality of frequency components extracted.
  • production and synthesis stages are implemented respectively by means of Fast Fourier Transform and Inverse Fourier transform.
  • the invention also provides two variants of combined echo cancellation and suppression system noise.
  • the frequency processing unit 100 includes a cascade extraction circuit energy components 10, an estimation circuit of signal to noise ratio 11, a gain calculation circuit 12 and a circuit for synthesizing filter coefficients 13.
  • the time processing circuit 14 is a time filter, of type finite impulse response filter (FIR Filter in Anglo-Saxon terminology for Finite Impulse Response Filter).
  • the sampled noisy speech signal s (nT) resulting from this sampling is applied, on the one hand, to an input of the energy component extraction circuit 10 in the frequency processing unit 100, and, on the other hand, at an input of the time filter 14.
  • FIG. 2 diagrams the processing carried out in circuit 10 which receives the noisy speech signal s (nT).
  • This sampled noisy speech signal s (nT) is in the form of successive frames of samples, four of them T (n-2), T (n-1), T (n) and T (n +1) being represented in a first line of FIG. 2.
  • K 3 blocks of samples B (1), B (2) and B (3) , K being an integer.
  • K 3 blocks of samples are formed according to the embodiment described from, on the one hand, this frame of rank n, T (n), and, on the other hand, the 2 frames T (n-2) and T (n-1), of respective ranks (n-2) and (n-1).
  • the step denoted 101 in this figure 2 aims to simplify the implementation of the subsequent treatments by selecting only part of the frequency components in each group E (1) i to E (3) i , 0 ⁇ i ⁇ 255.
  • This step is based on the following property.
  • the Fast Fourier Transform of a real signal has pseudo-symmetry.
  • These first 129 selected frequency components are sufficient to describe each group E (k) i , 0 ⁇ i ⁇ 255 , in a complete way since the other frequency components in the group, namely the last 127 components E (k) 129 to E ( k) 255, are deduced by symmetry.
  • the frequency components E (k) O to E (k) 128 which are selected in each group indeed present a character of symmetry compared to the complementary E (k) 129 to E (k) 255 of these components selected among all the components frequencies in the group that are produced initially.
  • the frequency components E (k) O to E (k) 128 are thus produced for each group.
  • the 129 frequency components selected in each group are decimated by 2, in order to retain only one component out of two from each group of components selected.
  • This decimation by two 102 aims at selectively removing one component out of two, relating to a given frequency, with a view to inhibiting the interaction effect produced on this component each of the two frequency components located at two respective frequencies on the one hand and d 'other of said given frequency.
  • the device 10 extracts for a cycle relating to a frame T (n) for processing the noisy speech signal s (nT), 65 energy components Em j , each representative of an energy, or power, of the noisy speech signal s (nT) for the frequency, or frequency band, considered.
  • all of the steps 100, 101 and 102 described in relation to FIG. 2, although improving the implementation of the method according to the invention, can be reduced to a single step consisting in the application of a single Fourier transform FFT at M 128 samples of the frame T (n) retained for the processing cycle considered.
  • the selection step 101 may either be operated or else not be operated, and this directly on the frequency components resulting from the FFT treatment (s).
  • the 65 energy components Emj, 0 ⁇ j ⁇ 64 are applied to an input of the signal-to-noise ratio estimation circuit 11.
  • circuit 11 estimates a signal-to-noise ratio SNR j between the noisy speech signal s (nT) and a noise signal included in this noisy speech signal, for the energy component considered Em j .
  • SNR j not Em j not / B j not in which the exponent n identifies the processing cycle relating to the frame T (n) of rank n and B j denotes a noise energy component in the energy component Em j of rank j.
  • such an estimation of the signal to noise ratio is based on the calculation of the noise energy component estimated in each given energy component. It uses for example the ratio between the energy component Em j n extracted and the noise energy component B j n-1 previously calculated during a processing cycle which preceded the considered processing cycle for suppressing the noise signal in the frame T (n). The higher this ratio, the more it translates the existence of a speech signal for the frequency energy component Em j n considered, in which case the noise component B j (n-1) calculated relative to the energy component Em j ( n-1) for the rank (n-1) is maintained in the noise component B j n .
  • the circuit 11 assigns a signal to noise ratio SNR j, 0 ⁇ j ⁇ 64 , to each extracted energy component Emj, 0 ⁇ j ⁇ 64 , according to an estimation algorithm using such a principle.
  • SNR j As a function of these 65 signal-to-noise ratios SNR j , the circuit 12 calculates for each of them a gain G j , taking for example a value substantially between 0 and 1, which is directly linked to the signal-to-noise ratio SNR j for the corresponding frequency component.
  • the noise signal component for each frequency energy component Em j is attenuated.
  • These gains G j are such that the weighting of the energy components Em j respectively by these gains would give a discrete spectrum of weighted frequency energy components, which would be representative of the noisy speech signal s (nT) in which the noise signal is substantially suppressed.
  • This circuit 13 comprises a first circuit (not shown) for duplicating the 65 calculated gains Gj, in accordance with the equation given in (1).
  • These coefficients denoted C (nT), 128 in number, are applied to a first control input of filter 14, typically FIR filter.
  • a second input of the filter 14 receives the noisy speech signal s (nT).
  • the filter 14 operates a convolution of the coefficients C (nT) with the 128 samples of the frame T (n), into a denoised frame of 128 samples forming part of the denoised speech signal s * (nT).
  • the method resulting from the device described above is naturally "adaptive" in the sense that the coefficients C (nT) applied to the control input of the FIR filter 14 are modified for each frame T (n), of rank n given, according to the processing 10, 11, 12 and 13 of the samples forming the speech signal to be processed.
  • the characteristic main process of noise suppression according to the invention is to use, on the one hand, a treatment digital frequency 100 of the noisy speech signal, for produce temporal filter coefficients C (nT), and, on the other hand, a digital temporal processing 14 of the signal speech noise s (nT) as a function of the coefficients of filtering C (nT), to produce a speech signal s * (nT) in which the noise signal is substantially suppressed.
  • a combined system of noise cancellation and echo cancellation which is included in a terminal, typically a hands-free radiotelephone, includes a microphone 2, a speaker 4, a device for noise suppression according to the invention 1, as described previously, a time processing circuit 14 'and a echo canceller 3.
  • the noise canceling device 1 is identical to the device shown in Figure 1, and mainly includes a frequency processing unit 100 and a time processing circuit 14.
  • the canceller echo consists of a subtractor 30 and a circuit 31 which produces an estimated echo signal.
  • Microphone 2 receives a speech signal to be transmitted [s (t) + e (t)] formed by a signal of noisy sound speech s (t) added to an echo signal and).
  • This echo signal results from the acoustic coupling between speaker 4 and microphone 2.
  • the device noise suppression 1 milking as described above, the speech signal to be transmitted as a speech signal transmitted denoised [s * (nT) + e * (nT)] which is applied to a first input of subtractor 30 including a second input receives the output of circuit 31.
  • a speech signal received r (t) from a remote terminal is applied, from a part to a speaker input, and, secondly, to a input of circuit 31 through the processing circuit temporal 14 'preceded by a sampler 14a'.
  • the circuit time processing 14 ' is strictly at all times similar to the time processing circuit 14 in the noise canceling device 1 (FIG.1).
  • This characteristic is based on the fact that the estimated echo of the received signal r (t) produced by circuit 31 is to be subtracted, in subtractor 30, to the echo signal processed by noise suppression e * (nT) in circuit 1 and not at original echo signal e (nT).
  • This circuit 14 ' is therefore a outright duplication of the time processing circuit 14 in device 1, as indicated by the arrow in broken line at two ends in Figure 3.
  • the time processing circuit 14 ′ is therefore associated with each instant at the same 128 filter coefficients C (nT) as the circuit 14 in device 1. It processes the signal from received speech r (t) into a noisy received speech signal r * (nT).
  • a combined system of noise cancellation and echo cancellation comprises a microphone 2, a speaker 4, an echo canceller 3, a signal unit frequency processing 100, a processing circuit temporal 14 and a sampler 5.
  • Units 100 and circuit 14 are identical to those described in FIG. 1.
  • the echo canceller 3 is formed by a subtractor 30 and a circuit 31 which produces an estimated echo signal ê (nT).
  • the microphone 2 receives a transmitted speech signal [s (t) + e (t)] formed by a noisy audible speech signal s (t) added to an echo signal e (t). This echo signal results from coupling acoustic between speaker 4 and microphone 2.
  • the transmitted speech signal [s (t) + e (t)] is sampled in the sampler 5 in the signal [s (nT) + e (nT)].
  • This signal sampled is applied, on the one hand, to an input of the unit 100, and, on the other hand, at an input of the circuit 14 to through subtractor 30.
  • a speech signal received in from a remote terminal r (t) - is applied, from a on the other hand, at an input of circuit 31, and, on the other hand, at a speaker input 4.
  • Circuit 31, receiving the signal r (t) produces in response an estimated echo signal ê (nT) applied to a first input of the subtractor 30, one of which second input receives the transmitted speech signal [s (nT) + e (nT)].
  • a difference signal [s (nT) + e (nT) -ê (nT)] applied to the circuit 14.
  • the treatments implemented in unit 100 operate on the transmitted speech signal [s (nT) + e (nT)]
  • the time processing in circuit 14 from coefficients C (nT) produced by the unit 100, operates on the difference signal, or processed speech signal processed by echo cancellation, [s (nT) + e (nT) -ê (nT)].
  • This variant avoids "duplicating" circuit 14 in the branch including circuit 31, as represented by the arrow in lines discontinuous in Figure 3 for the previous variant.

Abstract

The method involves applying noisy input speech signals (S(t)) to a sampler (1a). The signal is divided, with part of the signal passed to a noisy signal processor (100). The processor extracts noisy energy signals (10) and passes them to an S/N ratio estimator (11). The signal is then passed to a gain calculation circuit (12) before input to a circuit determining the filter function coefficient. The output C(nT) from the filter function circuit is applied to a FIR filter (14) in the second channel.

Description

La présente invention concerne de manière générale des procédé et dispositif de suppression d'un signal de bruit dans un signal de parole, typiquement pour une application à la radiotéléphonie main-libre. Elle a trait également à un système mettant en oeuvre un tel dispositif en combinaison avec un annuleur d'écho. The present invention relates generally to method and device for suppressing a noise signal in a speech signal, typically for an application to hands-free radiotelephony. It also relates to a system implementing such a device in combination with an echo canceller.

Dans un environnement bruité, le signal électrique résultant d'une conversion acousto-électrique d'un signal de parole est mélangé à un signal de bruit. Dès lors que le niveau du signal de bruit dans cet environnement est élevé, par exemple dans l'habitacle d'un véhicule, il est nécessaire de mettre en oeuvre des traitements visant à supprimer le signal de bruit dans le signal électrique de parole. Typiquement, deux types de traitement de suppression de bruit sont distingués selon la technique antérieure : le traitement par soustraction spectrale et le traitement dit par banc de filtres. In a noisy environment, the electrical signal resulting from an acousto-electric conversion of a signal speech is mixed with a noise signal. As soon as the noise level in this environment is high, for example in the passenger compartment of a vehicle, it is necessary to implement treatments aimed at suppress the noise signal in the electrical signal of speech. Typically, two types of removal processing are distinguished according to the prior art: the spectral subtraction processing and the so-called processing by filter bank.

Le traitement par banc de filtres, tel que décrit dans le brevet américain US-A-4 628 529, consiste en une étape de séparation du signal d'entrée en une pluralité de signaux temporels représentatifs chacun d'une bande de fréquences prédéterminée respective, une étape d'estimation d'un rapport signal sur bruit pour chacun de ces signaux temporels, une étape de pondération de ces signaux temporels par des coefficients respectifs qui sont chacun fonction de l'un respectif des rapports signal sur bruit pour le signal temporel considéré, et une étape d'addition de ces signaux temporels pondérés en un signal résultant consistant en un signal de parole dans lequel le signal de bruit est supprimé. Typiquement, chacun des rapports signal sur bruit est estimé en fonction de la variation de la puissance du signal temporel considéré dans sa bande de fréquences respective. Un tel traitement par banc de filtres nécessite des moyens de calcul considérables en raison du fait que toutes les étapes de séparation, estimation, pondération et addition précitées sont réalisées dans le domaine temporel. En pratique, dans un radiotéléphone, les moyens de calcul étant limités, en termes de Millions d'Instructions Par Seconde (Mips), par la capacité du processeur de traitement de signal numérique (DSP en terminologie anglo-saxonne pour Digital Signal Processor), il est alors proposé de réduire le traitement de suppression de signal de bruit à des bandes de fréquences grossières, et donc réduire la finesse, ou précision, de ce traitement. Treatment by filter bank, as described in American patent US-A-4 628 529, consists of a step of separating the input signal into a plurality of representative time signals each of a band of respective predetermined frequencies, an estimation step a signal-to-noise ratio for each of these signals temporal, a step of weighting these temporal signals by respective coefficients which are each a function of the respective one signal to noise ratios for the signal time considered, and a step of adding these signals time weighted as a resulting signal consisting of a speech signal in which the noise signal is deleted. Typically, each of the signal-to-noise ratios is estimated based on the variation of the power of the time signal considered in its frequency band respective. Such a treatment by filter bank requires   considerable means of calculation due to the fact that all stages of separation, estimation, weighting and above addition are carried out in the time domain. In practice, in a radiotelephone, the means of calculation being limited, in terms of Millions of Instructions Per Second (Mips), by the capacity of the processing processor digital signal (DSP in English terminology for Digital Signal Processor), it is then proposed to reduce noise signal suppression processing at bands coarse frequencies, and therefore reduce the fineness, or precision, of this treatment.

Le traitement par soustraction spectrale recourt pour sa part au domaine fréquentiel, typiquement par utilisation de transformée de Fourier Rapide FFT (Fast Fourier Transform en terminologie anglo-saxonne). Il présente l'inconvénient principal d'induire une distorsion non linéaire dans le signal de parole traité qui résulte de la perte d'information de phase de ce signal. En effet ce traitement par soustraction spectrale produit une telle distorsion car il applique aux échantillons résultant de la Transformée de Fourier Rapide du signal de parole bruité à traiter des fonctions de module au carré qui suppriment l'information de phase, rendant ainsi ce traitement non linéaire. En outre, le défaut de linéarité du traitement par soustraction spectrale empêche son utilisation efficacement en combinaison avec un traitement d'annulation d'écho, telle que proposée par l'invention, car le dispositif d'annulation d'écho est perturbé dans son fonctionnement par cette perte d'information de phase. Spectral subtraction processing uses its part in the frequency domain, typically by use Fast Fourier Transform (FFT) in Anglo-Saxon terminology). It has the disadvantage main to induce nonlinear distortion in the processed speech signal resulting from loss phase information of this signal. Indeed this treatment by spectral subtraction produces such a distortion because it applies to samples resulting from the Transformation of Fast Fourier of the noisy speech signal to process squared module functions that suppress information from phase, making this treatment non-linear. In addition, the lack of linearity of the subtraction processing spectral prevents its effective use in combination with echo cancellation processing, such as as proposed by the invention, because the cancellation device echo is disturbed in its functioning by this loss phase information.

Un premier objectif de la présente invention est de fournir un procédé de suppression de bruit dans un signal de parole qui a pour avantage de réduire considérablement la puissance de calcul nécessaire en termes de Nombre d'Instructions par Seconde, comparativement à un traitement par banc de filtres. A first objective of the present invention is to provide a method of suppressing noise in a signal speech which has the advantage of considerably reducing the computing power required in terms of Number Instructions Per Second, Compared To Treatment by filter bank.  

Un second objectif de l'invention est de fournir un procédé n'induisant pas une distorsion non linéaire dans le signal de parole à traiter, par contraste avec le traitement par soustraction spectrale. A second objective of the invention is to provide a process which does not induce non-linear distortion in the speech signal to be processed, in contrast to the processing by spectral subtraction.

Un autre objectif de l'invention est de fournir un système comprenant un dispositif de suppression de bruit mettant en oeuvre les étapes du procédé, en combinaison avec un dispositif d'annulation d'écho. Another object of the invention is to provide a system including a noise canceling device implementing the process steps, in combination with an echo canceller.

A cette fin, un procédé de suppression d'un signal de bruit dans un signal de parole bruité qui est échantillonné est caractérisé selon l'invention par les étapes de :

  • traitement fréquentiel numérique dudit signal de parole bruité, pour produire des coefficients temporels de filtrage, et
  • traitement temporel numérique dudit signal de parole bruité en fonction desdits coefficients de filtrage, en un signal de parole dans lequel ledit signal de bruit est sensiblement supprimé.
To this end, a method of suppressing a noise signal in a noisy speech signal which is sampled is characterized according to the invention by the steps of:
  • digital frequency processing of said noisy speech signal, to produce temporal filtering coefficients, and
  • digital time processing of said noisy speech signal as a function of said filter coefficients, into a speech signal in which said noise signal is substantially suppressed.

Le procédé définit, pour un cycle de traitement donné, les étapes de traitement fréquentiel numérique de:

  • extraction d'une pluralité de composantes énergétiques fréquentielles dans ledit signal de parole bruité,
  • estimation, pour chacune des composantes énergétiques fréquentielles extraites, d'un rapport entre un niveau d'énergie du signal de parole bruité et un niveau d'énergie du signal de bruit,
  • détermination d'un gain respectif pour ladite chacune des composantes énergétiques fréquentielles extraites, en fonction dudit rapport estimé entre le niveau d'énergie du signal de parole bruité et le niveau d'énergie du signal de bruit pour ladite chacune sélectionnée des composantes fréquentielles, et
  • synthèse desdits coefficients de filtrage en fonction desdits gains.
The method defines, for a given processing cycle, the digital frequency processing steps of:
  • extraction of a plurality of frequency energy components from said noisy speech signal,
  • estimation, for each of the frequency energy components extracted, of a relationship between an energy level of the noisy speech signal and an energy level of the noise signal,
  • determination of a respective gain for said each of the extracted frequency energy components, as a function of said estimated ratio between the energy level of the noisy speech signal and the energy level of the noise signal for said each of selected frequency components, and
  • synthesis of said filter coefficients as a function of said gains.

De préférence, l'étape d'extraction de composantes énergétiques fréquentielles comprend les sous-étapes de

  • production de K groupes comprenant chacun une pluralité de composantes fréquentielles, respectivement pour K blocs entrelacés du signal de parole bruité, avec K entier, et
  • calcul d'une moyenne énergétique de K composantes fréquentielles de même rang respectivement dans les K groupes, en l'une respective des composantes énergétiques fréquentielles extraites.
Preferably, the step of extracting frequency energy components comprises the sub-steps of
  • production of K groups each comprising a plurality of frequency components, respectively for K interleaved blocks of the noisy speech signal, with K integer, and
  • calculation of an energy average of K frequency components of the same rank respectively in the K groups, in a respective one of the frequency energy components extracted.

Typiquement, l'étape de calcul est précédée, pour chacun des K groupes de composantes fréquentielles, par une étape de sélection d'une partie des composantes fréquentielles ayant des rangs prédéterminés respectifs dans ledit chacun des groupes, ladite partie sélectionnée présentant un caractère de symétrie par rapport au complémentaire de cette partie parmi la pluralité des composantes fréquentielles extraites. Par ailleurs, les étapes de production et synthèse sont mises en oeuvre respectivement au moyen de Transformée de Fourier Rapide et Transformée de Fourier Inverse. Typically, the calculation step is preceded, for each of the K groups of frequency components, by a step of selecting a part of the components frequencies with respective predetermined ranks in said each of the groups, said selected part presenting a character of symmetry with respect to the complementary to this part among the plurality of frequency components extracted. In addition, production and synthesis stages are implemented respectively by means of Fast Fourier Transform and Inverse Fourier transform.

Un dispositif pour la mise en oeuvre du procédé comprend, pour chacun de cycles de traitement successifs,:

  • des moyens pour extraire une pluralité de composantes énergétiques fréquentielles dans ledit signal de parole bruité,
  • des moyens pour estimer, pour chacune des composantes énergétiques fréquentielles extraites, un rapport entre un niveau d'énergie du signal de parole bruité et un niveau d'énergie de signal de bruit,
  • des moyens pour déterminer un gain respectif pour ladite chacune des composantes énergétiques fréquentielles extraites, en fonction dudit rapport estimé entre le niveau d'énergie du signal de parole bruité et le niveau d'énergie du signal de bruit pour ladite chacune sélectionnée des composantes fréquentielles,
  • des moyens pour synthétiser lesdits coefficients de filtrage en fonction desdits gains, et
  • des moyens de filtrage temporel dudit signal de parole bruit en fonction desdits coefficients de filtrage, en un signal de parole dans lequel ledit signal de bruit est sensiblement supprimé.
A device for implementing the method comprises, for each of the successive treatment cycles:
  • means for extracting a plurality of frequency energy components from said noisy speech signal,
  • means for estimating, for each of the frequency energy components extracted, a relationship between an energy level of the noisy speech signal and a noise signal energy level,
  • means for determining a respective gain for said each of the extracted frequency energy components, as a function of said estimated ratio between the energy level of the noisy speech signal and the energy level of the noise signal for said selected each of the frequency components,
  • means for synthesizing said filter coefficients as a function of said gains, and
  • means for the temporal filtering of said noise speech signal as a function of said filtering coefficients, into a speech signal in which said noise signal is substantially suppressed.

L'invention fournit également deux variantes de système combiné d'annulation d'écho et de suppression de bruit. The invention also provides two variants of combined echo cancellation and suppression system noise.

Selon une première variante, le système comprend:

  • un dispositif de suppression de bruit pour supprimer un signal de bruit dans un signal de parole à transmettre, en un signal de parole débruité,
  • un annuleur d'écho comprenant un premier moyen pour produire un signal d'écho estimé en fonction d'un signal de parole donné et d'un signal de différence, et un second moyen pour soustraire ledit signal d'écho estimé audit signal de parole débruité, en ledit signal de différence.
According to a first variant, the system comprises:
  • a noise suppression device for suppressing a noise signal in a speech signal to be transmitted, into a noise-reduced speech signal,
  • an echo canceller comprising first means for producing an estimated echo signal based on a given speech signal and a difference signal, and second means for subtracting said estimated echo signal from said speech signal denoised, in said difference signal.

Il est caractérisé en ce que ledit dispositif de suppression de bruit est sous la forme de:

  • un moyen de traitement fréquentiel numérique dudit signal de parole à transmettre, pour produire des coefficients temporels de filtrage, et
  • un premier moyen de traitement temporel numérique pour traiter ledit signal de parole à transmettre en fonction desdits coefficients de filtrage, en ledit signal de parole débruité dans lequel ledit signal de bruit est sensiblement supprimé, et en ce que ledit système comprend en outre:
  • un second moyen de traitement temporel numérique, strictement similaire audit premier moyen de traitement temporel, pour traiter un signal de parole reçu d'un terminal distant en fonction desdits coefficients de filtrage, en ledit signal de parole donné.
It is characterized in that said noise suppression device is in the form of:
  • means for digital frequency processing of said speech signal to be transmitted, to produce temporal filter coefficients, and
  • first digital time processing means for processing said speech signal to be transmitted as a function of said filter coefficients, into said denoised speech signal in which said noise signal is substantially suppressed, and in that said system further comprises:
  • second digital time processing means, strictly similar to said first time processing means, for processing a speech signal received from a remote terminal according to said filter coefficients, into said given speech signal.

Selon une seconde variante, le système comprend

  • un annuleur d'écho comprenant un premier moyen pour produire un signal d'écho estimé en fonction d'un signal de parole reçu d'un terminal distant et d'un signal de différence, et un second moyen pour soustraire ledit signal d'écho estimé à un signal de parole à transmettre, en ledit signal de différence,
According to a second variant, the system comprises
  • an echo canceller comprising first means for producing an estimated echo signal based on a speech signal received from a remote terminal and a difference signal, and second means for subtracting said echo signal estimated at a speech signal to be transmitted, in said difference signal,

Il est caractérisé en ce qu'il comprend en outre:

  • un dispositif de suppression de bruit pour supprimer un signal de bruit dans le signal de différence, en un signal de parole débruité, ledit dispositif de suppression de bruit étant sous la forme de:
  • un moyen de traitement fréquentiel numérique pour traiter ledit signal de parole à transmettre, afin de produire des coefficients temporels de filtrage, et
  • un moyen de traitement temporel numérique pour traiter ledit signal de différence en fonction desdits coefficients de filtrage, en un signal de parole débruité dans lequel ledit signal de bruit est sensiblement supprimé.
It is characterized in that it further comprises:
  • a noise suppression device for suppressing a noise signal in the difference signal, into a noise-suppressed speech signal, said noise suppression device being in the form of:
  • digital frequency processing means for processing said speech signal to be transmitted, in order to produce temporal filter coefficients, and
  • digital time processing means for processing said difference signal as a function of said filter coefficients, into a denoised speech signal in which said noise signal is substantially suppressed.

D'autres caractéristiques et avantages de la présente invention apparaítront puis clairement à la lecture de la description suivante, en référence aux dessins annexés correspondants dans lesquels :

  • la figure 1 montre sous la forme d'un bloc-diagramme un dispositif selon l'invention de suppression de bruit dans un signal de parole;
  • la figure 2 schématise des étapes de traitement mises en oeuvre dans un circuit du dispositif de la figure 1;
  • la figure 3 montre, sous forme de bloc-diagramme, une première réalisation selon l'invention d'un système mettant en oeuvre le dispositif de la figure 1 en combinaison avec un annuleur d'écho; et
  • la figure 4 montre, sous forme de bloc-diagramme, une seconde réalisation selon l'invention d'un système mettant en oeuvre le dispositif de la figure 1 en combinaison avec un annuleur d'écho
Other characteristics and advantages of the present invention will appear then clearly on reading the following description, with reference to the corresponding appended drawings in which:
  • FIG. 1 shows in the form of a block diagram a device according to the invention for suppressing noise in a speech signal;
  • Figure 2 shows schematically the processing steps implemented in a circuit of the device of Figure 1;
  • FIG. 3 shows, in the form of a block diagram, a first embodiment according to the invention of a system implementing the device of FIG. 1 in combination with an echo canceller; and
  • FIG. 4 shows, in the form of a block diagram, a second embodiment according to the invention of a system implementing the device of FIG. 1 in combination with an echo canceller

En référence à la figure 1, un dispositif selon l'invention de suppression d'un signal de bruit dans un signal de parole 1 comprend un circuit d'échantillonnage la, une unité de traitement fréquentiel 100, et un circuit de traitement temporel 14. L'unité de traitement fréquentiel 100 comprend en cascade un circuit d'extraction de composantes énergétiques 10, un circuit d'estimation de rapport signal sur bruit 11, un circuit de calcul de gain 12 et un circuit de synthèse de coefficients de filtrage 13. Le circuit de traitement temporel 14 est un filtre temporel, de type filtre à réponse impulsionnelle finie (FIR Filter en terminologie anglo-saxonne pour Finite Impulse Response Filter). Referring to Figure 1, a device according to the invention of suppressing a noise signal in a speech signal 1 comprises a sampling circuit la, a frequency processing unit 100, and a circuit temporal processing 14. The frequency processing unit 100 includes a cascade extraction circuit energy components 10, an estimation circuit of signal to noise ratio 11, a gain calculation circuit 12 and a circuit for synthesizing filter coefficients 13. The time processing circuit 14 is a time filter, of type finite impulse response filter (FIR Filter in Anglo-Saxon terminology for Finite Impulse Response Filter).

Le circuit d'échantillonnage la échantillonne à une fréquence F=1/T un signal analogique de parole bruité s(t) constitué d'un signal de bruit additionné à un signal de parole. Le signal de parole bruité échantillonné s(nT) résultant de cet échantillonnage est appliqué, d'une part, à une entrée du circuit d'extraction de composantes énergétiques 10 dans l'unité de traitement fréquentiel 100, et, d'autre part, à une entrée du filtre temporel 14. La figure 2 schématise les traitements effectués dans le circuit 10 qui reçoit le signal de parole bruité s(nT). Ce signal de parole bruité échantillonné s(nT) se présente sous la forme de trames successives d'échantillons, quatre d'entr'elles T(n-2), T(n-1), T(n) et T(n+1) étant représentées à une première ligne de la figure 2. Une trame T(n) est formée, selon la réalisation décrite, par M = 128 échantillons, notés e(n)m, l'indice m variant entre O et 127. Pour chaque trame de rang n, T(n), associée à un cycle de traitement donné du procédé selon l'invention, sont produits K=3 blocs d'échantillons B(1), B(2) et B(3), K étant un nombre entier. Ces K=3 blocs d'échantillons sont formés selon la réalisation décrite à partir, d'une part, de cette trame de rang n, T(n), et, d'autre part, des 2 trames T(n-2) et T(n-1), de rangs respectifs (n-2) et (n-1). Les K=3 blocs d'échantillons B(1) à B(3) sont entrelacés et comprennent chacun 2.M= 256 échantillons successifs dans les trames T(n-2) à T(n), à compter de K=3 premiers échantillons respectifs de rangs 0 et M/2=64 dans la trame T(n-2) et de rang O dans la trame T(n-1). Les groupes respectifs de 2.M échantillons formant les blocs B(1), B(2) et B(3) sont notés b(1)i, b(2)i; et b(3)i, i variant de 0 à (2.M - 1)=255. A ces groupes respectifs d'échantillons b(1)i, b(2)i, b(3)i, 0 ≤i ≤255, sont appliquées respectivement trois transformées de Fourier rapides identiques selon les étapes notées 100a, 100b, et 100c. Ces étapes de transformée de Fourier rapide peuvent éventuellement être précédées par une opération de fenétrage temporel. Par ces transformées de Fourier rapides, à chacun des K=3 groupes d'échantillons b(1)i, b(2)i et b(3)i, est associé l'un respectif de K=3 groupes de composantes fréquentielles, notés E(1)i, E(2)i et E(3)i, l'indice i variant de 0 à 255. L'étape notée 101 dans cette figure 2 vise à simplifier la mise en oeuvre des traitements ultérieurs en sélectionnant seulement une partie des composantes fréquentielles dans chaque groupe E(1)i à E(3)i, 0 ≤i ≤255. Cette étape se fonde sur la propriété suivante. La Transformée de Fourier Rapide d'un signal réel présente une pseudo-symétrie. En effet, en sachant que les échantillons formant le signal de parole sont réels, chaque groupe de composantes fréquentielles E(k)i, k= 1, 2 ou 3, peut s'écrire sous la forme: E(k)i={E(k)0,E(k)1,.,E(k)127,E(k)128,E(k)129=E(k)127,.,E(k)255=E(k)1} L'étape de traitement 101 sélectionne, dans chaque groupe E(k=1)i ,E(k=2)i et E(k=3)i, 0 ≤i ≤255, une partie des composantes fréquentielles le constituant, à savoir les composantes fréquentielles E(k)O à E(k)₁₂₈ de rang 0 à 128, qui forment un groupe de composantes fréquentielles sélectionnées. Ces 129 premières composantes fréquentielles sélectionnées sont suffisantes pour décrire chaque groupe E(k)i, 0 ≤i ≤255, de façon complète puisque les autres composantes fréquentielles dans le groupe, à savoir les 127 dernières composantes E(k)₁₂₉ à E(k)255, se déduisent par symétrie. Les composantes fréquentielles E(k)O à E(k)₁₂₈ qui sont sélectionnées dans chaque groupe présentent en effet un caractère de symétrie par rapport au complémentaire E(k)₁₂₉ à E(k)₂₅₅ de ces composantes sélectionnées parmi toutes les composantes fréquentielles dans le groupe qui sont produites initialement. En sortie du traitement 101, sont ainsi produites les composantes fréquentielles E(k)O à E(k)₁₂₈ pour chaque groupe. Selon l'étape 102, les 129 composantes fréquentielles sélectionnées dans chaque groupe sont décimées par 2, en vue de ne retenir qu'une composante sur deux parmi chaque groupe de composantes sélectionné. Cette décimation par deux 102 vise à écarter sélectivement une composante sur deux, relative à une fréquence donnée, en vue d'inhiber l'effet d'interaction que produit sur cette composante chacune des deux composantes fréquentielles situées à deux fréquences respectives de part et d'autre de ladite fréquence donnée. En pratique, sont retenues les 65 composantes fréquentielles E(k)i, avec i=1, 3, 5,.., 127,128, sachant que la composante fréquentielle E(k)₀ ne présente aucun intérêt à être retenue puisqu'il s'agit d'une composante continue. A des fins de simplification de notation, ces composantes fréquentielles E(k)i, avec i=1, 3, 5,..,127,128 sont notées E(k)j, avec 0 ≤j ≤64. En résultat des étapes 101 et 102, est ainsi produite pour chaque groupe initial de composantes E(1)i ,E(2)i et E(3)i, 0 ≤i ≤255, un groupe de composantes sélectionnées et décimées. The sampling circuit samples it at a frequency F = 1 / T an analog noisy speech signal s (t) consisting of a noise signal added to a speech signal. The sampled noisy speech signal s (nT) resulting from this sampling is applied, on the one hand, to an input of the energy component extraction circuit 10 in the frequency processing unit 100, and, on the other hand, at an input of the time filter 14. FIG. 2 diagrams the processing carried out in circuit 10 which receives the noisy speech signal s (nT). This sampled noisy speech signal s (nT) is in the form of successive frames of samples, four of them T (n-2), T (n-1), T (n) and T (n +1) being represented in a first line of FIG. 2. A frame T (n) is formed, according to the embodiment described, by M = 128 samples, denoted e (n) m , the index m varying between O and 127 For each frame of rank n, T (n), associated with a given processing cycle of the method according to the invention, are produced K = 3 blocks of samples B (1), B (2) and B (3) , K being an integer. These K = 3 blocks of samples are formed according to the embodiment described from, on the one hand, this frame of rank n, T (n), and, on the other hand, the 2 frames T (n-2) and T (n-1), of respective ranks (n-2) and (n-1). The K = 3 blocks of samples B (1) to B (3) are interleaved and each comprise 2.M = 256 successive samples in the frames T (n-2) to T (n), from K = 3 first respective samples of ranks 0 and M / 2 = 64 in the frame T (n-2) and of rank O in the frame T (n-1). The respective groups of 2.M samples forming the blocks B (1), B (2) and B (3) are denoted b (1) i , b (2) i ; and b (3) i , i varying from 0 to (2.M - 1) = 255. To these respective groups of samples b (1) i , b (2) i , b (3) i , 0 ≤i ≤255 , three identical fast Fourier transforms are applied respectively according to the steps noted 100a, 100b, and 100c . These fast Fourier transform steps can possibly be preceded by a time window operation. By these fast Fourier transforms, to each of K = 3 groups of samples b (1) i , b (2) i and b (3) i , is associated the respective one of K = 3 groups of frequency components, denoted E (1) i , E (2) i and E (3) i , the index i varying from 0 to 255. The step denoted 101 in this figure 2 aims to simplify the implementation of the subsequent treatments by selecting only part of the frequency components in each group E (1) i to E (3) i , 0 ≤i ≤255. This step is based on the following property. The Fast Fourier Transform of a real signal has pseudo-symmetry. Indeed, knowing that the samples forming the speech signal are real, each group of frequency components E (k) i , k = 1, 2 or 3, can be written in the form: E (k) i = {E (k) 0 , E (k) 1 ,., E (k) 127 , E (k) 128 , E (k) 129 = E (k) 127 ,., E (k) 255 = E (k) 1 } The processing step 101 selects, in each group E (k = 1) i , E (k = 2) i and E (k = 3) i , 0 ≤i ≤255 , a part of the frequency components constituting it, with know the frequency components E (k) O to E (k) ₁₂₈ of rank 0 to 128, which form a group of selected frequency components. These first 129 selected frequency components are sufficient to describe each group E (k) i , 0 ≤i ≤255 , in a complete way since the other frequency components in the group, namely the last 127 components E (k) ₁₂₉ to E ( k) 255, are deduced by symmetry. The frequency components E (k) O to E (k) ₁₂₈ which are selected in each group indeed present a character of symmetry compared to the complementary E (k) ₁₂₉ to E (k) ₂₅₅ of these components selected among all the components frequencies in the group that are produced initially. At the end of the processing 101, the frequency components E (k) O to E (k) ₁₂₈ are thus produced for each group. According to step 102, the 129 frequency components selected in each group are decimated by 2, in order to retain only one component out of two from each group of components selected. This decimation by two 102 aims at selectively removing one component out of two, relating to a given frequency, with a view to inhibiting the interaction effect produced on this component each of the two frequency components located at two respective frequencies on the one hand and d 'other of said given frequency. In practice, the 65 frequency components E (k) i are retained, with i = 1, 3, 5, .., 127.128, knowing that the frequency component E (k) ₀ has no interest in being retained since it is a continuous component. For the purposes of simplification of notation, these frequency components E (k) i , with i = 1, 3, 5, .., 127.128 are denoted E (k) j , with 0 ≤j ≤64 . As a result of steps 101 and 102, a group of selected and decimated components is thus produced for each initial group of components E (1) i , E (2) i and E (3) i , 0 ≤i ≤255 .

Selon l'étape 103, il est opéré un calcul de la moyenne énergétique de chaque triplet de K=3 composantes fréquentielles de même rang j dans les K=3 groupes de composantes fréquentielles sélectionnées et décimées E(1)j, E(2)j et E(3)j, l'indice j variant de 0 à 64, pour produire 65 composantes énergétiques moyennées Emj, j variant de 0 à 64. Ce calcul comprend l'élévation au carré du module de chaque composante fréquentielle de même rang j dans les K=3 groupes de composantes sélectionnées et décimées, en K=3 composantes énergétiques, puis la moyenne de ces K=3 composantes énergétiques. According to step 103, a calculation is made of the energy average of each triplet of K = 3 frequency components of the same rank j in the K = 3 groups of frequency components selected and decimated E (1) j , E (2) j and E (3) j , the index j varying from 0 to 64, to produce 65 averaged energy components Em j , j varying from 0 to 64. This calculation includes the squared elevation of the module of each frequency component likewise rank j in the K = 3 groups of selected and decimated components, in K = 3 energy components, then the average of these K = 3 energy components.

Ainsi le dispositif 10 extrait pour un cycle relatif à une trame T(n) de traitement du signal de parole bruité s(nT), 65 composantes énergétiques Emj, chacune représentative d'une énergie, ou puissance, du signal de parole bruité s (nT) pour la fréquence, ou bande de fréquence, considérée. Il est à noter que toutes les étapes 100, 101 et 102 décrites relativement à la figure 2, bien qu'améliorant la mise en oeuvre du procédé selon l'invention, peuvent être réduites à une seule étape consistant en l'application d'une unique transformée de Fourier FFT aux M = 128 échantillons de la trame T(n) retenue pour le cycle de traitement considéré. En outre, l'étape de sélection 101 peut ou bien être opérée ou bien ne pas être opérée, et cela directement sur les composantes fréquentielles résultant du(des) traitement(s) FFT. Thus the device 10 extracts for a cycle relating to a frame T (n) for processing the noisy speech signal s (nT), 65 energy components Em j , each representative of an energy, or power, of the noisy speech signal s (nT) for the frequency, or frequency band, considered. It should be noted that all of the steps 100, 101 and 102 described in relation to FIG. 2, although improving the implementation of the method according to the invention, can be reduced to a single step consisting in the application of a single Fourier transform FFT at M = 128 samples of the frame T (n) retained for the processing cycle considered. In addition, the selection step 101 may either be operated or else not be operated, and this directly on the frequency components resulting from the FFT treatment (s).

En revenant à la figure 1, les 65 composantes énergétiques Emj, 0 ≤j ≤64, sont appliquées à une entrée du circuit d'estimation de rapport signal sur bruit 11. Pour chacune de ces 65 composantes énergétiques extraites Emj, le circuit 11 estime un rapport de signal sur bruit SNRj entre le signal de parole bruité s(nT) et un signal de bruit inclus dans ce signal de parole bruité, pour la composante énergétique considérée Emj. Un tel rapport signal sur bruit est donné par: SNRj n = Emj n/Bj n dans laquelle l'exposant n repère le cycle de traitement relatif à la trame T(n) de rang n et Bj dénote une composante d'énergie de bruit dans la composante énergétique Emj de rang j. Returning to FIG. 1, the 65 energy components Emj, 0 ≤j ≤64 , are applied to an input of the signal-to-noise ratio estimation circuit 11. For each of these 65 extracted energy components Em j , circuit 11 estimates a signal-to-noise ratio SNR j between the noisy speech signal s (nT) and a noise signal included in this noisy speech signal, for the energy component considered Em j . Such a signal-to-noise ratio is given by: SNR j not = Em j not / B j not in which the exponent n identifies the processing cycle relating to the frame T (n) of rank n and B j denotes a noise energy component in the energy component Em j of rank j.

En pratique, une telle estimation du rapport signal sur bruit se base sur le calcul de la composante d'énergie de bruit estimée dans chaque composante énergétique donnée. Elle utilise par exemple le rapport entre la composante énergétique Emj n extraite et la composante énergétique de bruit Bj n-1 calculée précédemment au cours d'un cycle de traitement ayant précédé le cycle considéré de traitement de suppression de signal de bruit dans la trame T(n). Plus ce rapport est élevé, plus il traduit l'existence d'un signal de parole pour la composante énergétique fréquentielle Emj n considérée, auquel cas la composante de bruit Bj (n-1) calculée relativement à la composante énergétique Emj (n-1) pour le rang (n-1) est maintenue en la composante de bruit Bj n . Plus ce rapport est faible, plus il traduit le fait que la composante énergétique se résume à un signal de bruit, auquel cas la composante de bruit Bj n varie par calcul en conséquence. Le circuit 11 attribue un rapport signal sur bruit SNRj, 0 ≤j ≤64, à chaque composante énergétique extraite Emj, 0 ≤j ≤64, selon un algorithme d'estimation utilisant un tel principe. En fonction de ces 65 rapports signal sur bruit SNRj, le circuit 12 calcule pour chacun d'eux, un gain Gj, prenant par exemple une valeur comprise sensiblement entre 0 et 1, qui est directement lié au rapport signal sur bruit SNRj pour la composante fréquentielle correspondante. Pour une composante énergétique fréquentielle donnée Emj, plus le rapport SNRj du signal de parole bruité s(nT) sur signal de bruit est élevé, plus le gain Gj est faible, et plus le rapport SNRj du signal de parole bruité sur le signal de bruit est faible, plus le gain Gj est élevé. Ainsi est atténuée la composante de signal de bruit pour chaque composante énergétique fréquentielle Emj. Ces gains Gj sont tels que la pondération des composantes énergétiques Emj respectivement par ces gains donnerait un spectre discret de composantes énergétiques fréquentielles pondérées, qui serait représentatif du signal de parole bruité s(nT) dans lequel le signal de bruit est sensiblement supprimé. In practice, such an estimation of the signal to noise ratio is based on the calculation of the noise energy component estimated in each given energy component. It uses for example the ratio between the energy component Em j n extracted and the noise energy component B j n-1 previously calculated during a processing cycle which preceded the considered processing cycle for suppressing the noise signal in the frame T (n). The higher this ratio, the more it translates the existence of a speech signal for the frequency energy component Em j n considered, in which case the noise component B j (n-1) calculated relative to the energy component Em j ( n-1) for the rank (n-1) is maintained in the noise component B j n . The lower this ratio, the more it reflects the fact that the energy component boils down to a noise signal, in which case the noise component B j n varies by calculation accordingly. The circuit 11 assigns a signal to noise ratio SNR j, 0 ≤j ≤64 , to each extracted energy component Emj, 0 ≤j ≤64 , according to an estimation algorithm using such a principle. As a function of these 65 signal-to-noise ratios SNR j , the circuit 12 calculates for each of them a gain G j , taking for example a value substantially between 0 and 1, which is directly linked to the signal-to-noise ratio SNR j for the corresponding frequency component. For a given frequency energy component Em j , the higher the ratio SNR j of the noisy speech signal s (nT) to the noise signal, the lower the gain G j , and the higher the ratio SNR j of the noisy speech signal on the lower the noise signal, the higher the gain G j . Thus the noise signal component for each frequency energy component Em j is attenuated. These gains G j are such that the weighting of the energy components Em j respectively by these gains would give a discrete spectrum of weighted frequency energy components, which would be representative of the noisy speech signal s (nT) in which the noise signal is substantially suppressed.

Une sortie du circuit 12 produisant les gains Gj est appliquée à une entrée du circuit de synthèse des coefficients de filtrage 13. Ce circuit 13 comprend un premier circuit (non représenté) de duplication des 65 gains calculés Gj, en conformité avec l'équation donnée en (1). Ce circuit reçoit 65 gains, notés G₀, G₁,..,G₆₄, et produit 128 gains pouvant s'écrire sous la forme d'un groupe de gains Gj, j compris entre 0 et 127, tel que suit: Gj={G0,G1,..,G63,G64,G65=G63,..,G127=G1} An output of the circuit 12 producing the gains G j is applied to an input of the circuit for synthesizing the filter coefficients 13. This circuit 13 comprises a first circuit (not shown) for duplicating the 65 calculated gains Gj, in accordance with the equation given in (1). This circuit receives 65 gains, noted G₀, G₁, .., G₆₄, and produces 128 gains which can be written in the form of a group of gains G j, j between 0 and 127, as follows: G j = {G 0 , G 1 , .., G 63 , G 64 , G 65 = G 63 , .., G 127 = G 1 }

Un second circuit (non représenté) dans le circuit de synthèse 13, sous la forme d'une Transformée de Fourier Inverse TFD⁻¹, synthétise 128 coefficients C(nT) du filtre 14 par transformée de Fourier inverse des 128 gains Gj. Ces coefficients notés C(nT), au nombre de 128, sont appliquées à une première entrée de commande du filtre 14, typiquement filtre FIR. Une seconde entrée du filtre 14 reçoit le signal de parole bruité s(nT). Le filtre 14 opère une convolution des coefficients C(nT) avec les 128 échantillons de la trame T(n), en une trame débruitée de 128 échantillons formant partie du signal de parole débruité s*(nT). Le procédé résultant du dispositif décrit ci-dessus est naturellement "adaptatif" en ce sens que les coefficients C(nT) appliqués à l'entrée de commande du filtre FIR 14 sont modifiés pour chaque trame T(n), de rang n donné, en fonction des traitements 10, 11, 12 et 13 des échantillons formant le signal de parole à traiter. A second circuit (not shown) in the synthesis circuit 13, in the form of an Inverse Fourier Transform TFD⁻¹, synthesizes 128 coefficients C (nT) of the filter 14 by inverse Fourier transform of the 128 gains G j . These coefficients denoted C (nT), 128 in number, are applied to a first control input of filter 14, typically FIR filter. A second input of the filter 14 receives the noisy speech signal s (nT). The filter 14 operates a convolution of the coefficients C (nT) with the 128 samples of the frame T (n), into a denoised frame of 128 samples forming part of the denoised speech signal s * (nT). The method resulting from the device described above is naturally "adaptive" in the sense that the coefficients C (nT) applied to the control input of the FIR filter 14 are modified for each frame T (n), of rank n given, according to the processing 10, 11, 12 and 13 of the samples forming the speech signal to be processed.

En résumé de ce qui précède, la caractéristique principale du procédé de suppression de bruit selon l'invention est d'utiliser, d'une part, un traitement fréquentiel numérique 100 du signal de parole bruité, pour produire des coefficients temporels de filtrage C(nT), et, d'autre part, un traitement temporel numérique 14 du signal de parole bruité s(nT) en fonction des coefficients de filtrage C(nT), pour produire un signal de parole s*(nT) dans lequel le signal de bruit est sensiblement supprimé. In summary from the above, the characteristic main process of noise suppression according to the invention is to use, on the one hand, a treatment digital frequency 100 of the noisy speech signal, for produce temporal filter coefficients C (nT), and, on the other hand, a digital temporal processing 14 of the signal speech noise s (nT) as a function of the coefficients of   filtering C (nT), to produce a speech signal s * (nT) in which the noise signal is substantially suppressed.

En référence à la figure 3, un système combiné de suppression de bruit et d'annulation d'écho selon une première variante de l'invention qui est inclus dans un terminal, typiquement un radiotéléphone main-libre, comprend un microphone 2, un haut parleur 4, un dispositif de suppression de bruit selon l'invention 1, tel que décrit précédemment, un circuit de traitement temporel 14' et un annuleur d'écho 3. Le dispositif de suppression de bruit 1 est identique au dispositif montré à la figure 1, et comprend principalement une unité de traitement fréquentiel 100 et un circuit de traitement temporel 14. L'annuleur d'écho est formé d'un soustracteur 30 et d'un circuit 31 qui produit un signal d'écho estimé. Le microphone 2 reçoit un signal de parole à transmettre [s(t)+e(t)] formé d'un signal de parole sonore bruité s(t) additionné à un signal d'écho e(t). Ce signal d'écho résulte du couplage acoustique entre les haut-parleur 4 et microphone 2. Le dispositif de suppression de bruit 1 traite, tel que décrit précédemment, le signal de parole à transmettre en un signal de parole transmis débruité [s*(nT)+e*(nT)] qui est appliqué à une première entrée du soustracteur 30 dont une seconde entrée reçoit la sortie du circuit 31. Un signal de parole reçu r(t) en provenance d'un terminal distant est appliqué, d'une part à une entrée du haut-parleur, et, d'autre part, à une entrée du circuit 31 à travers le circuit de traitement temporel 14' précédé par un échantillonneur 14a'. Selon une caractéristique importante de l'invention, le circuit de traitement temporel 14' est à chaque instant strictement similaire au circuit de traitement temporel 14 dans le dispositif de suppression de bruit 1 (FIG.1). Cette caractéristique se base sur le fait que l'écho estimé du signal reçu r(t) produit par le circuit 31 est à soustraire, dans le soustracteur 30, au signal d'écho traité par suppression de bruit e*(nT) dans le circuit 1 et non au signal d'écho original e(nT). Ce circuit 14' est donc une duplication pure et simple du circuit de traitement temporel 14 dans le dispositif 1, comme indiqué par la flèche en trait discontinu à deux extrémités dans la figure 3. Le circuit de traitement temporel 14' est donc associé à chaque instant aux mêmes 128 coefficients de filtrage C(nT) que le circuit 14 dans le dispositif 1. Il traite le signal de parole reçu r(t) en un signal de parole reçu débruité r*(nT). Ce traitement résulte de la convolution, par cycle de 128, des coefficients C(nT) et des échantillons r(nT) du signal reçu r(t). Le circuit 31 produit un signal d'estimation ê*(nT) du signal d'écho débruité e*(nT) à partir du signal de parole reçu débruité r*(nT) et de coefficients d'annulation d'écho w(nT). En sortie du soustracteur 30 est donc produit un signal de différence [s*(nT)+e*(nT)-ê*(nT)] dans lequel le signal d'écho est sensiblement supprimé. Les coefficients d'annulation d'écho w(nT) sont obtenus à partir de ce signal de différence. Referring to Figure 3, a combined system of noise cancellation and echo cancellation according to a first variant of the invention which is included in a terminal, typically a hands-free radiotelephone, includes a microphone 2, a speaker 4, a device for noise suppression according to the invention 1, as described previously, a time processing circuit 14 'and a echo canceller 3. The noise canceling device 1 is identical to the device shown in Figure 1, and mainly includes a frequency processing unit 100 and a time processing circuit 14. The canceller echo consists of a subtractor 30 and a circuit 31 which produces an estimated echo signal. Microphone 2 receives a speech signal to be transmitted [s (t) + e (t)] formed by a signal of noisy sound speech s (t) added to an echo signal and). This echo signal results from the acoustic coupling between speaker 4 and microphone 2. The device noise suppression 1 milking, as described above, the speech signal to be transmitted as a speech signal transmitted denoised [s * (nT) + e * (nT)] which is applied to a first input of subtractor 30 including a second input receives the output of circuit 31. A speech signal received r (t) from a remote terminal is applied, from a part to a speaker input, and, secondly, to a input of circuit 31 through the processing circuit temporal 14 'preceded by a sampler 14a'. According to one an important characteristic of the invention, the circuit time processing 14 'is strictly at all times similar to the time processing circuit 14 in the noise canceling device 1 (FIG.1). This characteristic is based on the fact that the estimated echo of the received signal r (t) produced by circuit 31 is to be subtracted, in subtractor 30, to the echo signal processed by noise suppression e * (nT) in circuit 1 and not at   original echo signal e (nT). This circuit 14 'is therefore a outright duplication of the time processing circuit 14 in device 1, as indicated by the arrow in broken line at two ends in Figure 3. The time processing circuit 14 ′ is therefore associated with each instant at the same 128 filter coefficients C (nT) as the circuit 14 in device 1. It processes the signal from received speech r (t) into a noisy received speech signal r * (nT). This treatment results from convolution, by cycle of 128, coefficients C (nT) and samples r (nT) of signal received r (t). Circuit 31 produces a signal estimation ê * (nT) of the noisy echo signal e * (nT) at from the noisy received speech signal r * (nT) and from echo cancellation coefficients w (nT). At the end of subtractor 30 is therefore produced a difference signal [s * (nT) + e * (nT) -ê * (nT)] in which the echo signal is significantly removed. The echo cancellation coefficients w (nT) are obtained from this difference signal.

En référence à la figure 4, un système combiné de suppression de bruit et d'annulation d'écho selon une seconde variante de l'invention comprend un microphone 2, un haut parleur 4, un annuleur d'écho 3, une unité de traitement fréquentiel 100, un circuit de traitement temporel 14 et un échantillonneur 5. Les unité 100 et circuit 14 sont identiques à ceux décrits dans la figure 1. L'annuleur d'écho 3 est formé d'un soustracteur 30 et d'un circuit 31 qui produit un signal d'écho estimé ê(nT). Le microphone 2 reçoit un signal de parole transmis [s(t)+e(t)] formé d'un signal de parole sonore bruité s(t) additionné à un signal d'écho e(t). Ce signal d'écho résulte du couplage acoustique entre les haut-parleur 4 et microphone 2. Le signal de parole transmis [s(t)+e(t)] est échantillonné dans l'échantillonneur 5 en le signal [s(nT)+e(nT)]. Ce signal échantillonné est appliqué, d'une part, à une entrée de l'unité 100, et, d'autre part, à une entrée du circuit 14 à travers le soustracteur 30. Un signal de parole reçu en provenance d'un terminal distant r(t)- est appliqué, d'une part, à une entrée du circuit 31, et, d'autre part, à une entrée du haut-parleur 4. Le circuit 31, recevant le signal r(t), produit en réponse un signal d'écho estimé ê(nT) appliqué à une première entrée du soustracteur 30 dont une seconde entrée reçoit le signal de parole transmis [s(nT)+e(nT)]. En sortie du soustracteur 30, est produit un signal de différence [s(nT)+e(nT)-ê(nT)] appliqué au circuit 14. Selon cette variante, d'une part les traitements fréquentiels mis en oeuvre dans l'unité 100 s'opèrent sur le signal de parole transmis [s(nT)+e(nT)], et d'autre part le traitement temporel dans le circuit 14, à partir des coefficients C(nT) produits par l'unité 100, s'opère sur le signal de différence, ou signal de parole transmis traité par annulation d'écho, [s(nT)+e(nT)-ê(nT)]. Cette variante évite de "dupliquer" le circuit 14 dans la branche incluant le circuit 31, comme représenté par la flèche en trait discontinu dans la figure 3 pour la variante précédente. Referring to Figure 4, a combined system of noise cancellation and echo cancellation according to a second variant of the invention comprises a microphone 2, a speaker 4, an echo canceller 3, a signal unit frequency processing 100, a processing circuit temporal 14 and a sampler 5. Units 100 and circuit 14 are identical to those described in FIG. 1. The echo canceller 3 is formed by a subtractor 30 and a circuit 31 which produces an estimated echo signal ê (nT). The microphone 2 receives a transmitted speech signal [s (t) + e (t)] formed by a noisy audible speech signal s (t) added to an echo signal e (t). This echo signal results from coupling acoustic between speaker 4 and microphone 2. The transmitted speech signal [s (t) + e (t)] is sampled in the sampler 5 in the signal [s (nT) + e (nT)]. This signal sampled is applied, on the one hand, to an input of the unit 100, and, on the other hand, at an input of the circuit 14 to through subtractor 30. A speech signal received in   from a remote terminal r (t) - is applied, from a on the other hand, at an input of circuit 31, and, on the other hand, at a speaker input 4. Circuit 31, receiving the signal r (t), produces in response an estimated echo signal ê (nT) applied to a first input of the subtractor 30, one of which second input receives the transmitted speech signal [s (nT) + e (nT)]. At the output of the subtractor 30, is produced a difference signal [s (nT) + e (nT) -ê (nT)] applied to the circuit 14. According to this variant, on the one hand the treatments implemented in unit 100 operate on the transmitted speech signal [s (nT) + e (nT)], and on the other hand the time processing in circuit 14, from coefficients C (nT) produced by the unit 100, operates on the difference signal, or processed speech signal processed by echo cancellation, [s (nT) + e (nT) -ê (nT)]. This variant avoids "duplicating" circuit 14 in the branch including circuit 31, as represented by the arrow in lines discontinuous in Figure 3 for the previous variant.

Claims (8)

Procédé de suppression d'un signal de bruit dans un signal de parole bruité (s(nT)) qui est échantillonné, caractérisé par les étapes de : traitement fréquentiel numérique (100) dudit signal de parole bruité, pour produire des coefficients temporels de filtrage (C(nT)), et traitement temporel numérique (14) dudit signal de parole bruité (s(nT)) en fonction desdits coefficients de filtrage (C(nT)), en un signal de parole (s*(nT)) dans lequel ledit signal de bruit est sensiblement supprimé. Method for suppressing a noise signal in a noisy speech signal (s (nT)) which is sampled, characterized by the steps of: digital frequency processing (100) of said noisy speech signal, to produce temporal filter coefficients (C (nT)), and digital time processing (14) of said noisy speech signal (s (nT)) as a function of said filter coefficients (C (nT)), into a speech signal (s * (nT)) in which said noise signal is substantially deleted. Procédé de suppression d'un signal de bruit dans un signal de parole bruité conforme à la revendication 1, caractérisé, pour un cycle de traitement donné, par les étapes de traitement fréquentiel numérique de: extraction (10) d'une pluralité de composantes énergétiques fréquentielles (Emj) dans ledit signal de parole bruité (s(nT)), estimation (11), pour chacune des composantes énergétiques fréquentielles extraites, d'un rapport (SNRj) entre un niveau d'énergie du signal de parole bruité (s(nT)) et un niveau d'énergie du signal de bruit, détermination (12) d'un gain (Gj) respectif pour ladite chacune des composantes énergétiques fréquentielles extraites (Emj), en fonction dudit rapport (SNRj) estimé entre le niveau d'énergie du signal de parole bruité (s(nT)) et le niveau d'énergie du signal de bruit pour ladite chacune sélectionnée des composantes fréquentielles, et synthèse (13) desdits coefficients de filtrage (C(nT)) en fonction desdits gains (Gj). Method for suppressing a noise signal in a noisy speech signal according to claim 1, characterized, for a given processing cycle, by the steps of digital frequency processing of: extraction (10) of a plurality of frequency energy components (Em j ) in said noisy speech signal (s (nT)), estimation (11), for each of the frequency energy components extracted, of a ratio (SNR j ) between an energy level of the noisy speech signal (s (nT)) and an energy level of the noise signal, determination (12) of a respective gain (G j ) for said each of the extracted frequency energy components (Em j ), as a function of said ratio (SNR j ) estimated between the energy level of the noisy speech signal (s (nT) )) and the energy level of the noise signal for said each selected frequency components, and synthesis (13) of said filter coefficients (C (nT)) as a function of said gains (G j ). Procédé conforme à l'une quelconque des revendications 2 à 3, caractérisé en ce que ladite étape d'extraction de composantes énergétiques fréquentielles comprend les sous-étapes de production (100a, 100b, 100c) de K groupes comprenant chacun une pluralité de composantes fréquentielles (E(1)i, E(2)i, E(3)i), respectivement pour K blocs (B(1), B(2), B(3)) entrelacés du signal de parole bruité (s(nT)), avec K entier, et calcul (103) d'une moyenne énergétique de K composantes fréquentielles de même rang (j) respectivement dans les K groupes, en l'une respective des composantes énergétiques fréquentielles extraites. Process according to any one of Claims 2 to 3, characterized in that the said step of extracting frequency energy components comprises the substeps of production (100a, 100b, 100c) of K groups each comprising a plurality of frequency components (E (1) i , E (2) i , E (3) i ), respectively for K blocks (B (1), B ( 2), B (3)) interleaved with the noisy speech signal (s (nT)), with integer K, and calculation (103) of an energy average of K frequency components of the same rank (j) respectively in the K groups, in a respective one of the frequency energy components extracted. Procédé conforme à la revendication 3, caractérisé en ce que ladite étape de calcul (103) est précédée, pour chacun des K groupes de composantes fréquentielles, par une étape de sélection (101) d'une partie des composantes fréquentielles ayant des rangs prédéterminés respectifs dans ledit chacun des groupes (E(1)i, E(2)i, E(3)i), ladite partie sélectionnée présentant un caractère de symétrie par rapport au complémentaire de cette partie parmi la pluralité des composantes fréquentielles extraites. Method according to Claim 3, characterized in that the said calculation step (103) is preceded, for each of the K groups of frequency components, by a selection step (101) of a part of the frequency components having respective predetermined ranks in said each of the groups (E (1) i , E (2) i , E (3) i ), said selected part having a character of symmetry with respect to the complement of this part among the plurality of frequency components extracted. Procédé conforme à l'une quelconque des revendications 2 à 4, caractérisé en ce que lesdites étapes de production (100a, 100b, 100c) et de synthèse (13) sont mises en oeuvre respectivement au moyen de Transformée de Fourier Rapide et Transformée de Fourier Inverse. Process according to any of claims 2 to 4, characterized in that said steps production (100a, 100b, 100c) and synthesis (13) are implemented respectively by means of Transform Fast Fourier and Reverse Fourier Transform. Dispositif (1) de suppression d'un signal de bruit dans un signal de parole bruité (s(nT)) qui est échantillonné, caractérisé en ce qu'il comprend, pour chacun de cycles de traitement successifs,: des moyens (10) pour extraire une pluralité de composantes énergétiques fréquentielles (Emj) dans ledit signal de parole bruité (s(nT)), des moyens (11) pour estimer, pour chacune des composantes énergétiques fréquentielles extraites, un rapport (SNRj) entre un niveau d'énergie du signal de parole bruité (s(nT)) et un niveau d'énergie de signal de bruit, des moyens pour déterminer (12) un gain (Gj) respectif pour ladite chacune des composantes énergétiques fréquentielles extraites (Emj), en fonction dudit rapport (SNRj) estimé entre le niveau d'énergie du signal de parole bruité (s(nT)) et le niveau d'énergie du signal de bruit pour ladite chacune sélectionnée des composantes fréquentielles, des moyens (13) pour synthétiser lesdits coefficients de filtrage (C(nT)) en fonction desdits gains (Gj), et des moyens de filtrage temporel (14) dudit signal de parole bruité (s(nT)) en fonction desdits coefficients de filtrage (C(nT)), en un signal de parole (s(nT)) dans lequel ledit signal de bruit est sensiblement supprimé. Device (1) for suppressing a noise signal in a noisy speech signal (s (nT)) which is sampled, characterized in that it comprises, for each of successive processing cycles: means (10) for extracting a plurality of frequency energy components (Em j ) from said noisy speech signal (s (nT)), means (11) for estimating, for each of the frequency energy components extracted, a ratio (SNR j ) between an energy level of the noisy speech signal (s (nT)) and a noise signal energy level, means for determining (12) a respective gain (G j ) for said each of the extracted frequency energy components (Em j ), as a function of said ratio (SNR j ) estimated between the energy level of the noisy speech signal (s ( nT)) and the energy level of the noise signal for said each selected frequency components, means (13) for synthesizing said filter coefficients (C (nT)) as a function of said gains (G j ), and means for temporal filtering (14) of said noisy speech signal (s (nT)) as a function of said filter coefficients (C (nT)), into a speech signal (s (nT)) in which said noise signal is significantly removed. Système combiné d'annulation d'écho (3) et de suppression de bruit (1), comprenant un dispositif de suppression de bruit (1) pour supprimer un signal de bruit dans un signal de parole à transmettre (s(nT)+e(nT)), en un signal de parole débruité, un annuleur d'écho (3) comprenant un premier moyen (31) pour produire un signal d'écho estimé (ê*(t)) en fonction d'un signal de parole donné (r*(nT)) et d'un signal de différence (s*(nT)+e*(nT)-ê*(nT)), et un second moyen (30) pour soustraire ledit signal d'écho estimé (ê*(t)) audit signal de parole débruité (s*(nT)+e*(nT)), en ledit signal de différence (s*(nT)+e*(nT)-ê*(nT)),
caractérisé en ce que ledit dispositif de suppression de bruit est sous la forme de: un moyen de traitement fréquentiel numérique (100) dudit signal de parole à transmettre, pour produire des coefficients temporels de filtrage (C(nT)), et un premier moyen de traitement temporel numérique (14) pour traiter ledit signal de parole à transmettre (s(nT)) en fonction desdits coefficients de filtrage (C(nT)), en ledit signal de parole débruité (s*(nT)+e*(nT)) dans lequel ledit signal de bruit est sensiblement supprimé, et en ce que ledit système comprend en outre: un second moyen de traitement temporel numérique (14'), strictement similaire audit premier moyen de traitement temporel (14), pour traiter un signal de parole reçu d'un terminal distant (r(t)) en fonction desdits coefficients de filtrage (C(nT)), en ledit signal de parole donné (r*(nT)). Combined echo cancellation (3) and noise cancellation (1) system, comprising a noise suppression device (1) for suppressing a noise signal in a speech signal to be transmitted (s (nT) + e (nT)), into a noise-reduced speech signal, an echo canceller (3) comprising first means (31) for producing an estimated echo signal (ê * (t)) based on a given speech signal (r * (nT)) and a difference signal (s * (nT) + e * (nT) -ê * (nT)), and second means (30) for subtracting said estimated echo signal (ê * (t)) from said denoised speech signal (s * (nT) + e * (nT)), in said difference signal (s * (nT) + e * (nT) -ê * (nT)),
characterized in that said noise suppression device is in the form of: a digital frequency processing means (100) of said speech signal to be transmitted, to produce temporal filter coefficients (C (nT)), and first digital time processing means (14) for processing said speech signal to be transmitted (s (nT)) as a function of said filter coefficients (C (nT)), in said denoised speech signal (s * (nT) + e * (nT)) in which said noise signal is substantially suppressed, and in that said system further comprises: second digital time processing means (14 '), strictly similar to said first time processing means (14), for processing a speech signal received from a remote terminal (r (t)) as a function of said filter coefficients (C (nT)), in said given speech signal (r * (nT)). Système combiné d'annulation d'écho (3) et de suppression de bruit (1) dans un signal de parole à transmettre (s(nT)+e(nT)), comprenant un annuleur d'écho (3) comprenant un premier moyen (31) pour produire un signal d'écho estimé (ê(t)) en fonction d'un signal de parole (r(t)) reçu d'un terminal distant et d'un signal de différence (s(nT)+e(nT)-ê(nT)), et un second moyen (30) pour soustraire ledit signal d'écho estimé (ê(t)) à un signal de parole à transmettre (s(nT)+e(nT)), en ledit signal de différence (s(nT)+e(nT)-ê(nT)),
caractérisé en ce qu'il comprend: un dispositif de suppression de bruit (1) pour supprimer un signal de bruit dans le signal de différence (s(nT)+e(nT)-ê(nT)), en un signal de parole débruité (s*(nT)+e*(nT)-ê*(nT)), ledit dispositif de suppression de bruit étant sous la forme de: un moyen de traitement fréquentiel numérique (100) pour traiter ledit signal de parole à transmettre (s(nT)+e(nT)), afin de produire des coefficients temporels de filtrage (C(nT)), et un moyen de traitement temporel numérique (14) pour traiter ledit signal de différence (s(nT)+e(nT)-ê(nT)) en fonction desdits coefficients de filtrage (C(nT)), en un signal de parole débruité (s*(nT)+e*(nT)-ê*(nT)) dans lequel ledit signal de bruit est sensiblement supprimé. Combined echo cancellation (3) and noise cancellation (1) system in a speech signal to be transmitted (s (nT) + e (nT)), comprising an echo canceller (3) comprising first means (31) for producing an estimated echo signal (ê (t)) as a function of a speech signal (r (t)) received from a remote terminal and a difference signal (s (nT) + e (nT) -ê (nT)), and second means (30) for subtracting said estimated echo signal (ê (t)) from a speech signal at transmit (s (nT) + e (nT)), in said difference signal (s (nT) + e (nT) -ê (nT)),
characterized in that it comprises: a noise suppression device (1) for suppressing a noise signal in the difference signal (s (nT) + e (nT) -ê (nT)), into a noise-reduced speech signal (s * (nT) + e * (nT) -ê * (nT)), said noise suppression device being in the form of: digital frequency processing means (100) for processing said speech signal to be transmitted (s (nT) + e (nT)), in order to produce temporal filter coefficients (C (nT)), and digital time processing means (14) for processing said difference signal (s (nT) + e (nT) -ê (nT)) as a function of said filter coefficients (C (nT)), into a denoised speech signal (s * (nT) + e * (nT) -ê * (nT)) in which said noise signal is substantially suppressed.
EP95402385A 1994-10-28 1995-10-25 Method and apparatus for noise suppression in a speech signal and corresponding system with echo cancellation Expired - Lifetime EP0710947B1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
FR9412964A FR2726392B1 (en) 1994-10-28 1994-10-28 METHOD AND APPARATUS FOR SUPPRESSING NOISE IN A SPEAKING SIGNAL, AND SYSTEM WITH CORRESPONDING ECHO CANCELLATION
FR9412964 1994-10-28

Publications (2)

Publication Number Publication Date
EP0710947A1 true EP0710947A1 (en) 1996-05-08
EP0710947B1 EP0710947B1 (en) 2003-01-08

Family

ID=9468340

Family Applications (1)

Application Number Title Priority Date Filing Date
EP95402385A Expired - Lifetime EP0710947B1 (en) 1994-10-28 1995-10-25 Method and apparatus for noise suppression in a speech signal and corresponding system with echo cancellation

Country Status (10)

Country Link
US (1) US5680393A (en)
EP (1) EP0710947B1 (en)
JP (2) JPH08213936A (en)
AT (1) ATE230890T1 (en)
AU (1) AU698081B2 (en)
CA (1) CA2161575A1 (en)
DE (1) DE69529328T2 (en)
FI (1) FI955086A (en)
FR (1) FR2726392B1 (en)
NZ (1) NZ280224A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0884886A2 (en) * 1997-06-11 1998-12-16 Oki Electric Industry Co., Ltd. Echo canceler employing multiple step gains
CN104702245A (en) * 2015-02-04 2015-06-10 航天科工深圳（集团）有限公司 Method for suppressing interference of surging lightning waves

Families Citing this family (44)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2729804B1 (en) * 1995-01-24 1997-04-04 Matra Communication ACOUSTIC ECHO CANCELLER WITH ADAPTIVE FILTER AND PASSAGE IN THE FREQUENTIAL DOMAIN
JP2921472B2 (en) * 1996-03-15 1999-07-19 日本電気株式会社 Voice and noise elimination device, voice recognition device
US6192087B1 (en) 1996-11-15 2001-02-20 Conexant Systems, Inc. Method and apparatus for spectral shaping in signal-point limited transmission systems
US6278744B1 (en) 1996-11-15 2001-08-21 Conexant Systems, Inc. System for controlling and shaping the spectrum and redundancy of signal-point limited transmission
US5933495A (en) * 1997-02-07 1999-08-03 Texas Instruments Incorporated Subband acoustic noise suppression
US7853024B2 (en) * 1997-08-14 2010-12-14 Silentium Ltd. Active noise control system and method
IL121555A (en) * 1997-08-14 2008-07-08 Silentium Ltd Active acoustic noise reduction system
US6144735A (en) * 1998-03-12 2000-11-07 Westell Technologies, Inc. Filters for a digital subscriber line system for voice communication over a telephone line
US6115466A (en) * 1998-03-12 2000-09-05 Westell Technologies, Inc. Subscriber line system having a dual-mode filter for voice communications over a telephone line
US6717991B1 (en) * 1998-05-27 2004-04-06 Telefonaktiebolaget Lm Ericsson (Publ) System and method for dual microphone signal noise reduction using spectral subtraction
US6549586B2 (en) * 1999-04-12 2003-04-15 Telefonaktiebolaget L M Ericsson System and method for dual microphone signal noise reduction using spectral subtraction
US6122610A (en) * 1998-09-23 2000-09-19 Verance Corporation Noise suppression for low bitrate speech coder
US6507623B1 (en) * 1999-04-12 2003-01-14 Telefonaktiebolaget Lm Ericsson (Publ) Signal noise reduction by time-domain spectral subtraction
US6487257B1 (en) * 1999-04-12 2002-11-26 Telefonaktiebolaget L M Ericsson Signal noise reduction by time-domain spectral subtraction using fixed filters
US6411656B1 (en) * 1999-04-30 2002-06-25 3Com Corporation Echo cancelling softmodem
DE19925046A1 (en) * 1999-06-01 2001-05-03 Alcatel Sa Method and device for suppressing noise and echoes
US6137880A (en) * 1999-08-27 2000-10-24 Westell Technologies, Inc. Passive splitter filter for digital subscriber line voice communication for complex impedance terminations
US7225001B1 (en) 2000-04-24 2007-05-29 Telefonaktiebolaget Lm Ericsson (Publ) System and method for distributed noise suppression
DE10043064B4 (en) * 2000-09-01 2004-07-08 Dietmar Dr. Ruwisch Method and device for eliminating loudspeaker interference from microphone signals
FR2820227B1 (en) * 2001-01-30 2003-04-18 France Telecom NOISE REDUCTION METHOD AND DEVICE
US6963760B2 (en) * 2001-10-01 2005-11-08 General Motors Corporation Method and apparatus for generating DTMF tones using voice-recognition commands during hands-free communication in a vehicle
JP2004061617A (en) * 2002-07-25 2004-02-26 Fujitsu Ltd Received speech processing apparatus
KR20050026320A (en) * 2003-09-09 2005-03-15 삼성전자주식회사 Device and method for data reproduction
US7162212B2 (en) * 2003-09-22 2007-01-09 Agere Systems Inc. System and method for obscuring unwanted ambient noise and handset and central office equipment incorporating the same
US7065206B2 (en) * 2003-11-20 2006-06-20 Motorola, Inc. Method and apparatus for adaptive echo and noise control
US20070033030A1 (en) * 2005-07-19 2007-02-08 Oded Gottesman Techniques for measurement, adaptation, and setup of an audio communication system
US20080285767A1 (en) * 2005-10-25 2008-11-20 Harry Bachmann Method for the Estimation of a Useful Signal with the Aid of an Adaptive Process
US7774396B2 (en) * 2005-11-18 2010-08-10 Dynamic Hearing Pty Ltd Method and device for low delay processing
WO2007095664A1 (en) * 2006-02-21 2007-08-30 Dynamic Hearing Pty Ltd Method and device for low delay processing
JP4827661B2 (en) * 2006-08-30 2011-11-30 富士通株式会社 Signal processing method and apparatus
WO2008090544A2 (en) * 2007-01-22 2008-07-31 Silentium Ltd. Quiet fan incorporating active noise control (anc)
US7515703B1 (en) 2008-05-19 2009-04-07 International Business Machines Corporation Method and system for determining conference call embellishment tones and transmission of same
JP5056617B2 (en) * 2008-06-26 2012-10-24 沖電気工業株式会社 Background noise suppression / echo canceling apparatus, method and program
KR101737824B1 (en) * 2009-12-16 2017-05-19 삼성전자주식회사 Method and Apparatus for removing a noise signal from input signal in a noisy environment
JP5672437B2 (en) * 2010-09-14 2015-02-18 カシオ計算機株式会社 Noise suppression device, noise suppression method and program
US9431001B2 (en) 2011-05-11 2016-08-30 Silentium Ltd. Device, system and method of noise control
US9928824B2 (en) 2011-05-11 2018-03-27 Silentium Ltd. Apparatus, system and method of controlling noise within a noise-controlled volume
KR101306868B1 (en) * 2011-12-21 2013-09-10 황정진 Un-identified system modeling method and audio system for howling cancelation using it
US8878678B2 (en) * 2012-05-29 2014-11-04 Cisco Technology, Inc. Method and apparatus for providing an intelligent mute status reminder for an active speaker in a conference
JP2014085609A (en) * 2012-10-26 2014-05-12 Sony Corp Signal processor, signal processing method, and program
CN103795473B (en) * 2012-11-02 2017-04-12 华为技术有限公司 Method and system for eliminating power-frequency interference
CN103248337A (en) * 2013-05-29 2013-08-14 中国电子科技集团公司第二十二研究所 Spatio-temporal cascaded method for suppressing external interference of oblique ionogram
US11488615B2 (en) 2018-05-21 2022-11-01 International Business Machines Corporation Real-time assessment of call quality
JP7348812B2 (en) 2019-10-31 2023-09-21 東京都公立大学法人 Noise suppression device, noise suppression method, and voice input device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4628529A (en) 1985-07-01 1986-12-09 Motorola, Inc. Noise suppression system
EP0438174A2 (en) * 1990-01-18 1991-07-24 Matsushita Electric Industrial Co., Ltd. Signal processing device
EP0556992A1 (en) * 1992-02-14 1993-08-25 Nokia Mobile Phones Ltd. Noise attenuation system

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS6343451A (en) * 1986-08-11 1988-02-24 Mitsubishi Electric Corp Amplified speaking circuit
JPS63108822A (en) * 1986-10-25 1988-05-13 Nippon Telegr & Teleph Corp <Ntt> Noise suppressor for voice speech signal
US5561667A (en) * 1991-06-21 1996-10-01 Gerlach; Karl R. Systolic multiple channel band-partitioned noise canceller
JPH0563609A (en) * 1991-09-05 1993-03-12 Fujitsu Ltd Echo canceler system
US5416847A (en) * 1993-02-12 1995-05-16 The Walt Disney Company Multi-band, digital audio noise filter
US5329587A (en) * 1993-03-12 1994-07-12 At&T Bell Laboratories Low-delay subband adaptive filter
US5406622A (en) * 1993-09-02 1995-04-11 At&T Corp. Outbound noise cancellation for telephonic handset

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4628529A (en) 1985-07-01 1986-12-09 Motorola, Inc. Noise suppression system
EP0438174A2 (en) * 1990-01-18 1991-07-24 Matsushita Electric Industrial Co., Ltd. Signal processing device
EP0556992A1 (en) * 1992-02-14 1993-08-25 Nokia Mobile Phones Ltd. Noise attenuation system

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
F.M.WANG ET AL.: "Frequency domain adaptive postfiltering for enhancement of noisy speech", SPEECH COMMUNICATION, vol. 12, no. 1, AMSTERDAM, pages 41 - 56, XP000382195, DOI: doi:10.1016/0167-6393(93)90017-F *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0884886A2 (en) * 1997-06-11 1998-12-16 Oki Electric Industry Co., Ltd. Echo canceler employing multiple step gains
EP0884886A3 (en) * 1997-06-11 1999-08-04 Oki Electric Industry Co., Ltd. Echo canceler employing multiple step gains
US6236725B1 (en) 1997-06-11 2001-05-22 Oki Electric Industry Co., Ltd. Echo canceler employing multiple step gains
US6351532B1 (en) 1997-06-11 2002-02-26 Oki Electric Industry Co., Ltd. Echo canceler employing multiple step gains
CN104702245A (en) * 2015-02-04 2015-06-10 航天科工深圳（集团）有限公司 Method for suppressing interference of surging lightning waves

Also Published As

Publication number Publication date
CA2161575A1 (en) 1996-04-29
EP0710947B1 (en) 2003-01-08
FI955086A (en) 1996-04-29
JPH08213936A (en) 1996-08-20
US5680393A (en) 1997-10-21
AU698081B2 (en) 1998-10-22
JP4567655B2 (en) 2010-10-20
FR2726392A1 (en) 1996-05-03
DE69529328T2 (en) 2003-09-04
JP2007129736A (en) 2007-05-24
FI955086A0 (en) 1995-10-25
AU3444295A (en) 1996-05-09
NZ280224A (en) 1997-02-24
DE69529328D1 (en) 2003-02-13
ATE230890T1 (en) 2003-01-15
FR2726392B1 (en) 1997-01-10

Similar Documents

Publication Publication Date Title
EP0710947B1 (en) Method and apparatus for noise suppression in a speech signal and corresponding system with echo cancellation
EP0748555B1 (en) Acoustic echo suppressor with subband filtering
EP0103257B1 (en) Method and apparatus to reduce static noise
CA2266654C (en) Method and device for blind equalizing of transmission channel effects on a digital speech signal
EP0752181B1 (en) Frequency-domain adaptive-filter acoustic echo canceller
FR2789823A1 (en) UNILATERAL SUB-BAND FILTERS
EP0998166A1 (en) Device for audio processing,receiver and method for filtering the wanted signal and reproducing it in presence of ambient noise
EP0692883B1 (en) Blind equalisation method, and its application to speech recognition
EP1039736B1 (en) Method and device for adaptive identification and related adaptive echo canceller
EP0884926A1 (en) Method and device for optimized processing of an interfering signal when recording sound
FR3009121A1 (en) METHOD OF SUPPRESSING LATE REVERBERATION OF A SOUND SIGNAL
EP1103138B1 (en) Device for digital processing with frequency filtering and reduced computation complexity
JP2000134138A (en) Non-delay system to cancel subband echo
EP0059133B1 (en) Chrominance-signal de-emphasis filter and television receiver comprising such a filter
EP1438815A1 (en) Inter-symbol interference canceller
EP0824798B1 (en) Adaptive sub-band filtering
FR2678453A1 (en) Telephone instrument with echo-cancellation device
FR2771543A1 (en) Noise reduction algorithm
EP0848867A1 (en) Adaptive digital filtering process in the frequency domain

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FR GB IT LI NL SE

17P Request for examination filed

Effective date: 19961011

17Q First examination report despatched

Effective date: 19991230

RIC1 Information provided on ipc code assigned before grant

Free format text: 7G 10L 21/02 A

GRAG Despatch of communication of intention to grant

Free format text: ORIGINAL CODE: EPIDOS AGRA

RIC1 Information provided on ipc code assigned before grant

Free format text: 7G 10L 21/02 A

GRAG Despatch of communication of intention to grant

Free format text: ORIGINAL CODE: EPIDOS AGRA

GRAH Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOS IGRA

GRAH Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOS IGRA

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: ALCATEL

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AT BE CH DE DK ES FR GB IT LI NL SE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20030108

Ref country code: AT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20030108

REF Corresponds to:

Ref document number: 230890

Country of ref document: AT

Date of ref document: 20030115

Kind code of ref document: T

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

Free format text: NOT ENGLISH

REG Reference to a national code

Ref country code: CH

Ref legal event code: EP

REF Corresponds to:

Ref document number: 69529328

Country of ref document: DE

Date of ref document: 20030213

Kind code of ref document: P

GBT Gb: translation of ep patent filed (gb section 77(6)(a)/1977)

Effective date: 20030311

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20030408

REG Reference to a national code

Ref country code: SE

Ref legal event code: TRGR

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: ES

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20030730

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: LI

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20031031

Ref country code: CH

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20031031

Ref country code: BE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20031031

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed

Effective date: 20031009

BERE Be: lapsed

Owner name: *ALCATEL

Effective date: 20031031

REG Reference to a national code

Ref country code: CH

Ref legal event code: PL

REG Reference to a national code

Ref country code: FR

Ref legal event code: CD

REG Reference to a national code

Ref country code: GB

Ref legal event code: 732E

Free format text: REGISTERED BETWEEN 20131114 AND 20131120

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: SE

Payment date: 20131022

Year of fee payment: 19

Ref country code: GB

Payment date: 20131021

Year of fee payment: 19

Ref country code: DE

Payment date: 20131021

Year of fee payment: 19

Ref country code: FR

Payment date: 20131022

Year of fee payment: 19

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: IT

Payment date: 20131028

Year of fee payment: 19

REG Reference to a national code

Ref country code: FR

Ref legal event code: GC

Effective date: 20140717

REG Reference to a national code

Ref country code: FR

Ref legal event code: RG

Effective date: 20141016

REG Reference to a national code

Ref country code: DE

Ref legal event code: R119

Ref document number: 69529328

Country of ref document: DE

REG Reference to a national code

Ref country code: SE

Ref legal event code: EUG

GBPC Gb: european patent ceased through non-payment of renewal fee

Effective date: 20141025

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: GB

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20141025

Ref country code: SE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20141026

Ref country code: DE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20150501

REG Reference to a national code

Ref country code: FR

Ref legal event code: ST

Effective date: 20150630

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: FR

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20141031

Ref country code: IT

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20141025