Asynchronous HMM with applications to speech recognition

Abstract

We develop a novel formalism for modeling speech signals which are irregularly or incompletely sampled. This situation can arise in real world applications where the speech signal is being transmitted over an error prone channel where parts of the signal can be dropped. Typical speech systems based on Hidden Markov Models, cannot handle such data since HMMs rely on the assumption that observations are complete and made at regular intervals. In this paper we introduce the asynchronous HMM, a variant of the inhomogenous HMM commonly used in Bioinformatics, and show how it can be used to model irregularly or incompletely sampled data. A nested EM algorithm is presented in brief which can be used to learn the parameters of this asynchronous HMM. Evaluation on real world speech data that has been modified to simulate channel errors, shows that this model and its variants significantly outperforms the standard HMM and methods based on data interpolation.