New Approaches Towards Robust and Adaptive Speech Recognition
Hervé A. Bourlard, Samy Bengio, Katrin Weber · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · 2000
. In this paper, we discuss some new research directions in automatic speech recognition (ASR), and which somewhat deviate from the usual approaches. More specically, we will motivate and briey describe new approaches based on multi-stream and multi/band ASR. These approaches extend the standard hidden Markov model (HMM) based approach by assuming that the dierent (frequency) channels representing the speech signal are processed by dierent (independent) \\experts", each expert focusing on a dierent characteristic of the signal, and that the dierent stream likelihoods (or posteriors) are combined at some (temporal) stage to yield a global recognition output. As a further extension to multi-stream ASR, we will nally introduce a new approach, referred to as HMM2, where the HMM emission probabilities are estimated via state specic feature based HMMs responsible for merging the stream information and modeling their possible correlation. 2 IDIAP{RR 01-01 1 Multi-Channel Processing in...