Qiang Huo
Towards Robust Automatic Speech Recognition:
A Bayesian Perspective
Abstract:In this report, we start with a revisit to the statistical formulation of the automatic
speech recognition (ASR) problem, identify the factors which might influence the performance of the conventional plug-in MAP decision rule for ASR. We summarize our recent
research efforts on a class of robust speech recognition problem in which mismatches between training and testing conditions exist but an accurate knowledge of the mismatch
mechanism is unknown. The only available information is the test data along with a
set of pre-trained speech models and the decision parameters. We focus on two types of
Bayesian techniques, namely on-line Bayesian adaptation of hidden Markov model parameters and Bayesian predictive classification approach. We conclude the report with
a brief mention of our ongoing research efforts towards a robust and intelligent spoken
dialogue system.