An approach to robust speaker recognition using stochastic matching

This paper describes an approach to robust speaker recognition using stochastic matching for depressing channel mismatch in telephone networks. For the two recognition algorithms such as GMM (Gaussian mixture models) for text-independent speaker identification and HMM for text prompted speaker verification, the corresponding stochastic matching approaches are presented in detail. Experimental results have shown that the stochastic matching approach for channel compensation based on the compensation term from GMM is better than that of the compensation term from a single Gaussian model