This paper presents an adaptive algorithm for compensating pronunciation variations in hidden Markov model (HMM) based speech recognition. The proposed method aims to adapt the HMM topology and the corresponding HMM parameters to meet the variations of speaker dialects. In adaptive HMM topology, two hypothesis test schemes are designed to detect whether a new speaking variation occurs in state/phone levels. The test statistics are approximated by the chi-square densities. A new HMM topology is automatically generated by a significance level. Simultaneously, the HMM parameters and their hyperparameters are updated by Bayesian learning of the newly-generated Markov models. The pronunciation variations are coped with by a dialect adaptive HMM topology. We develop the incremental algorithm for corrective training of HMM topology and parameters. Experiments on TIMIT database show that the proposed algorithm is substantially better than the standard HMM with comparable size of parameters.
Bibliographic reference. Ting, Chuan-Wei / Lee, Kuo-Yuan / Chien, Jen-Tzung (2008): "Adaptive HMM topology for speech recognition", In INTERSPEECH-2008, 1237-1240.