No abstract

No abstract

No abstract

Recently, Bayesian learning has been developedas a mathematical tool for obtaining maximum a posteriori (MAP) estimates of hidden Markov model (HMM) parameters. This framework offers a way to incorporate newly acquired application-specific data into existing acoustic HMMs and combine them in an optimal manner. It is therefore an efficient technique for handling the sparse training data problem which is often encountered in speech recognition. In this study a number of important issues related to the application of Bayesian learning techniques to speaker adaptation are investigated. We show that the seed models required to construct prior densities to obtain the MAP estimate can be a speaker-independent model, a set of female and male models or even a task independent acoustic model. Speaker-adaptive training algorithms are shown to be effective in improving the performance of both speaker-dependent and speaker-independent speech recognition systems.

Abstruct-A minimax approach for robust classification of parametric information sources is studied and applied to isolatedword speech recognition based on hidden Markov modeling. The goal is to reduce the sensitivity of speech recognition systems to a possible mismatch between the training and testing conditions. To this end, a generalized likelihood ratio test is developed and shown to be optimal in the sense of achieving the highest asymptotic exponential rate of decay of the error probability for the worst-case mismatch situation. The proposed approach is compared to the standard approach, where no mismatch is assumed, in recognition of noisy speech and in other realistic mismatch situations.