International Workshop on Hands-Free Speech Communication (HSC2001) April 9-11, 2001 Kyoto, Japan

Evaluation of PMC for Segmental Unit Input HMM in Various Environments

Kazumasa Yamamoto (1), Seiichi Nakagawa (2), and Hiroshi Matsumoto (1)

(1) Faculty of Engineering, Shinshu University, Wakasato, Nagano, Japan
(2) Toyohashi University of Technology, Tenpaku, Toyohashi, Japan

For robust speech recognition in noisy environments, various methods have been studied. In this paper, we expanded the parallel model combination (PMC) for the segmental unit input HMM to recognize corrupted speech in additive noise and/or reverberant environments. Since Karhunen-Loeve expansion or LDA is used to reduce dimensionality of feature parameters in the segmental unit input HMM, the inverse transformation of segmental statistics to cepstral domain is needed. Then, the technique of original PMC can be used for remaining process. Experimental results showed the PMC for segmental unit input HMM proposed here gives better recognition performance than the original PMC in the additive noise environments. In the reverberant environments, however, the PMC for segment is not so effective.

Full Paper

Bibliographic reference.  Yamamoto, Kazumasa / Nakagawa, Seiichi / Matsumoto, Hiroshi (2001): "Evaluation of PMC for segmental unit input HMM in various environments", In HSC2001, 183-186.