Discriminative GMM-HMM acoustic model selection using two-level bayesian ying-yang harmony learning

  • Authors:

  • Zaihu Pang;Shikui Tu;Xihong Wu;Lei Xu

  • Affiliations:

  • Speech and Hearing Research Center, Key Laboratory of Machine Perception (Ministry of Education), Peking University, Beijing, China;Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, China;Speech and Hearing Research Center, Key Laboratory of Machine Perception (Ministry of Education), Peking University, Beijing, China,College of Computer Science and Technology, Jilin University, Ch ...;Speech and Hearing Research Center, Key Laboratory of Machine Perception (Ministry of Education), Peking University, Beijing, China,Department of Computer Science and Engineering, The Chinese Univ ...

  • Venue:
  • IScIDE'12 Proceedings of the third Sino-foreign-interchange conference on Intelligent Science and Intelligent Data Engineering
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a two-level Bayesian Ying-Yang (BYY) harmony learning based acoustic model discriminative training method. In this method, a rival penalized competitive learning (RPCL) simplified BYY harmony learning based discriminative training is conducted at the HMM state level to optimizing the state boundaries, while a BYY based model selection is conducted at the Gaussian mixture components level to determine the Gaussian mixture components within the same HMM state. Two levels of learning work coordinately and have good convergence. Experiments show that the trained model is more discriminative with better recognition performance, and also more compact with smaller number of Gaussian components.