International Workshop on Hands-Free Speech Communication (HSC2001) April 9-11, 2001 Kyoto, Japan

Robust Automatic Speech Recognition in Reverberant Environments by Model Selection

Laurent Couvreur and Christophe Couvreur

Signal Processing Department, Facult6 Polytechnique de Mons, Belgium Corporate R&D, Lernout & Hauspie, Belgium

This paper presents a method for robust automatic speech recognition (ASR) in reverberant environments. Our approach consists in the selection during operation of an acoustic model out of a library of models trained in various reverberaut conditions. The best model is selected by blindly estimating the full-band reverberation time. The estimation procedure is entirely based on the short-term log-energy sequence of the utterance to be recognized. Speech recognition experiments in simulated and real reverberant environments show the efficiency of our approach which outperforms standard channel normalization techniques.

Full Paper

Bibliographic reference.  Couvreur, Laurent / Couvreur, Christophe (2001): "Robust automatic speech recognition in reverberant environments by model selection", In HSC2001, 147-150.