HMMs with mixtures of trend functions for automatic speech recognition

In this study we extend the nonstationary-state (trended) HMM from the single-trend formulation of Deng (see Signal Processing, vo1.27, no.1, p. 65-78, 1992) to the mixture-trend one. This extension is motivated by the observation of wide variations in the trajectories of the acoustic data in fluent, speaker-independent speech associated with a given underlying linguistic unit. We show how HMMs with mixtures of trend functions can be implemented simply in the already well established singly trended HMM framework via the device of expanding each state into a set of parallel states. Details of a maximum-likelihood based algorithm are given for estimating state-dependent mixture trajectory parameters in the model. Experimental results on the task of classifying speaker-independent vowels excised from TIMIT database demonstrate consistent performance improvement using phonemic mixture-trended HMMs over their singly-trended counterpart.<
>