A Markov random field approach to Bayesian speaker adaptation

Speaker adaptation through Bayesian learning methodology is studied. In order to utilize the cross allophone correlations, a Markov random field model is proposed as the joint prior distribution of the allophones´ parameters. Neighborhoods are defined as pairs of parameters between which strong correlations have been observed previously. Maximum a posteriori estimates of the allophones´ mean vectors are obtained iteratively. This process is similar to a recursive prediction of the parameters. Further Bayesian smoothing is carried out by utilizing some simplifications on the functional forms of the marginal posterior distributions. Experimental results show rapid improvement of recognition accuracy