Continuously variable duration hidden Markov models for automatic speech recognition
Computer Speech and Language
A segmental HMM for speech pattern modelling
ICASSP'93 Proceedings of the 1993 IEEE international conference on Acoustics, speech, and signal processing: speech processing - Volume II
Hi-index | 0.89 |
A widely used method for parameterizing hidden semi-Markov model is using Gaussian distribution to form the output probability and using Gamma distribution to form the state duration probability. Most of these models are based on the classical Newton@?s method with second-order convergence, whose iterative convergence speed is slow for large-scale realtime applications. An improved parameter re-estimation algorithm is introduced for such parametric hidden semi-Markov model in this paper. The proposed approach is based on forward and backward algorithm. It applies an iterative method with eighth-order convergence to improve the performance of the model. The numerical examples validate the proposed method.