Constructing shared-state hidden Markov models based on a Bayesian approach.

摘要: In this paper, we propose a method for constructing sharedstate triphone HMMs (SST-HMMs) within practical Bayesian framework. our method, model selection criterion is derived SST-HMM based on the Variational approach. The appropriate phonetic decision tree structure of found by using according to given data set. This criterion, unlike conventional MDL applicable even in case insufficient amounts data. We conduct two experiments speaker independent word recognition order prove effectiveness proposed method. first experiment demonstrates that approach valid determining structure. second can design SST-HMMs with higher performance than when dealing small