Application of Variational Bayesian Approach to Speech Recognition
作者： Naonori Ueda , Shinji Watanabe , Atsushi Nakamura , Yasuhiro Minami
DOI:
关键词:
摘要: In this paper, we propose a Bayesian framework, which constructs shared-state triphone HMMs based on variational approach, and recognizes speech the prediction classification; estimation clustering for recognition (VBEC). An appropriate model structure with high performance can be found within VBEC framework. Unlike conventional methods, including BIC or MDL criterion maximum likelihood proposed selection is valid in principle, even when there are insufficient amounts of data, because it does not use an asymptotic assumption. isolated word experiments, show advantage over especially dealing small data.
nips.cc
elsevier.com参考文章(6)
Naonori Ueda, Shinji Watanabe, Atsushi Nakamura, Yasuhiro Minami, Constructing shared-state hidden Markov models based on a Bayesian approach. conference of the international speech communication association. ,(2002)
Koichi Shinoda, Takao Watanabe, Acoustic modeling based on the MDL principle for speech recognition. conference of the international speech communication association. ,(1997)
Naonori Ueda, Zoubin Ghahramani, None, Bayesian model search for mixture models based on optimizing variational bounds Neural Networks. ,vol. 15, pp. 1223- 1241 ,(2002) , 10.1016/S0893-6080(02)00040-0
Wu Chou, W. Reichl, Decision tree state tying based on penalized Bayesian information criterion international conference on acoustics speech and signal processing. ,vol. 1, pp. 345- 348 ,(1999) , 10.1109/ICASSP.1999.758133
Matthew J Beal, Z Ghahramani, None, Variational Inference for Bayesian Mixtures of Factor Analysers neural information processing systems. ,vol. 12, pp. 449- 455 ,(1999)
Hagai Attias, A Variational Baysian Framework for Graphical Models neural information processing systems. ,vol. 12, pp. 209- 215 ,(1999)