Structural MAP speaker adaptation using hierarchical priors

K. Shinoda, Chin-Hui Lee
1997 IEEE Workshop on Automatic Speech Recognition and Understanding Proceedings  
Most adaptation methods for speech recognition using hidden Markov models fall into two categories; one is the Bayesian approach, where prior distributions for the model parameters are assumed, and the other is the transformation based approach, where a predetermined simple transformation form is employed to modify the model parameters. It is known that the former is better when the amount of data for a d a p tation is large, while the latter is better when the amount of data is small. In this
more » ... is small. In this paper, we propose a new approach, structural maximum a posteriori (SMAP) approach, in which hierarchical priors are introduced to combine the two approaches above. The experimental results showed SMAP achieved better recognition accuracy than the two approaches for both small and large amounts of adaptation data.
doi:10.1109/asru.1997.659114 fatcat:pfpbmj7wv5ed3pheivgmd7bqda