ITRW on Non-Linear Speech Processing (NOLISP 03) May 20-23, 2003 Le Croisic, France

Blind Source Separation of FIR Convolutive Mixtures: Application to Speech Signals

Cédric Févotte, Alexandra Debiolles, Christian Doncarli

Institut de Recherche en Communications et Cybernétique de Nantes, France

In this paper we present a simple method to deal with Blind Source Separation (BSS) of Finite Impulse Response (FIR) convolutive mixtures. The global method proceeds in two steps. The first step consists in separating each source contribution in the mixture. This step provides several filtered version of each source. The second step consists in retrieving the original sources from the set of filtered versions of each source using a blind system identification method. We present some results on a mixture of speech and music.

Full Paper

Bibliographic reference.  Févotte, Cédric / Debiolles, Alexandra / Doncarli, Christian (2003): "Blind source separation of FIR convolutive mixtures: application to speech signals", In NOLISP-2003, paper 023.