Fixed-point algorithms for convolutive blind source separation based on non-gaussianity maximization Algorithmes à point xe pour séparation aveugle de sources convolutive fondée sur la maximisation de non-gaussianité

نویسندگان

  • Johan THOMAS
  • Yannick DEVILLE
  • Shahram HOSSEINI
چکیده

This paper presents a new approach to the problem of blind separation of independent components in the case of MA convolutive mixtures of MA processes. It consists of an extension of the well-known Fast-ICA algorithm developed by Hyvärinen and Oja for instantaneous mixtures. We introduce a new type of sphering (convolutive sphering) that allows the use of non-gaussianity criteria and associated parameter-free fast xed-point algorithms for the estimation of the source innovation processes. We prove the relevance of these criteria by reformulating the mixtures as linear instantaneous ones. We then describe associated kurtotic and negentropic time-domain algorithms. Test results are presented for arti cial coloured signals and for speech signals. Résumé Cet article présente une nouvelle approche pour la séparation aveugle de composantes indépendantes dans le cas des mélanges convolutifs MA de processus MA. Cette méthode peut être considérée comme une extension de l'algorithme Fast-ICA développé par Hyvärinen et Oja pour les mélanges instantanés. Nous introduisons un nouveau type de blanchiment ( sphering convolutif) qui permet l'utilisation de critères de non gaussianité associés à des algorithmes rapides de type point xe sans paramètre à ajuster, a n d'estimer les processus d'innovation des sources. Nous prouvons la pertinence de ces critères en reformulant le mélange sous forme instantanée. Nous dérivons ensuite les algorithmes à base de kurtosis et de négentropie qui en découlent. Des résultats de test sont présentés pour des signaux colorés arti ciels et pour des signaux de parole.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Algorithmes temporels rapides de type point fixe pour la séparation aveugle de mélanges convolutifs Time-domain fast fixed-point algorithms for blind separation of convolutive mixtures

This paper presents new blind separation methods for Moving Average (MA) convolutive mixtures of independent MA processes. They consist of time-domain extensions of the FastICA algorithms developed by Hyvärinen and Oja for instantaneous mixtures. They perform a convolutive sphering in order to use parameter-free fast fixed-point algorithms associated with kurtotic or negentropic nongaussianity ...

متن کامل

Multichannel nonnegative matrix factorization in convolutive mixtures for audio source separation Factorisation en matrices à coefficients positifs de données multicanal convolutives pour la séparation de sources audio

We consider inference in a general data-driven object-based model of multichannel audio data, assumed generated as a possibly underdetermined convolutive mixture of source signals. We work in the Short-Time Fourier Transform (STFT) domain, where convolution is routinely approximated as linear instantaneous mixing in each frequency band. Each source STFT is given a model inspired from nonnegativ...

متن کامل

A tractable framework for estimating and combining spectral source models for audio source separation

The underdetermined blind audio source separation (BSS) problem is often addressed in the time-frequency (TF) domain assuming that each TF point is modeled as an independent random variable with sparse distribution. On the other hand, methods based on structured spectral model, such as the Spectral Gaussian Scale Mixture Models (Spectral-GSMMs) or Spectral Nonnegative Matrix Factorization model...

متن کامل

Méthodes séquentielles de séparation aveugle de sources en présence de bruit

The topic which is addressed here concerns the blind separation of a multi-input/multi-output convolutive mixture of mutually independent sources. The considered framework includes the case of non necessarily i.i.d. sources, emphasis being put on iterative source separation methods (i.e. the sources are extracted one by one). More precisely, this work shows how to take into account the presence...

متن کامل

Séparation aveugle d’un mélange convolutif de sources non linéaires par une approche hiérarchique

This paper deals with blind source separation of convolutive mixtures of non linear (and hence non i.i.d.) sources. This general framework, although rarely treated in literature, is of great importance in applications. An iterative method was proposed [10]; however, the deflation procedure induces an error accumulation effect. We propose a hierarchical approach inspired from [6] which is less s...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2007