Online adaptive learning of continuous-density hidden markov models based on multiple-stream prior evolution and posterior pooling

We introduce a new adaptive Bayesian learning framework, called multiple-st ream prior evolution and posterior pooling, for online adaptation of the co ntinuous density hidden Markov model (CDHMM) parameters. Among three archit ectures we proposed for this framework, we study in detail a specific two-s tream system where linear transformations are applied to the mean vectors o f CDHMMs to control the evolution of their prior distribution, This new str eam of prior distribution can be combined with another stream of prior dist ribution evolved without any constraints applied, In a series of speaker ad aptation experiments on the task of continuous Mandarin speech recognition, we show that the new adaptation algorithm achieves a similar fast-adaptati on performance as that of the incremental maximum likelihood linear regress ion (MLLR) in the case of small amount of adaptation data, while maintains the good asymptotic convergence property as that of our previously proposed quasi-Bayes adaptation algorithms.