NOTE: The following materials are presented for timely dissemination of academic and technical work. Copyright and all other rights therein are reserved by authors and/or other copyright holders. Persoanl use of the following materials is permitted and, however, people using the materials or information are expected to adhere to the terms and constraints invoked by the related copyright.

Text-Dependent Speaker Identification Based on Input/Output HMMs: An Empirical Study


In this paper, we explore the Input/Output HMM (IOHMM) architecture for a substantial problem, that of text-dependent speaker identification. For subnetworks modeled with generalized linear models, we extend the IRLS algorithm to the M-step of the corresponding EM algorithm. Experimental results show that the improved EM algorithm yields significantly faster training than the original one. In comparison with the multilayer perceptron, the dynamic programming technique and hidden Markov models, we empirically demonstrate that the IOHMM architecture is a promising way to text-dependent speaker identification.

Click npl96.pdf for full text