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

Complimentary Combination of Microphone Array and HMM Composition for Noisy Speech Recognition

Takanobu Nishiura (1,2), Kazuhiro Miki (2), Satoshi Nakamura (1), Kiyohiro Shikano (2)

(1) ATR Spoken Language Translation Research Laboratories, Kyoto, Japan
(2) Graduate School of Information Science, Nara Institute of Science and Technology, Japan

Distant-talking speech recognition is very important in providing a natural interface for machints like self-moving robots. Distant-talking speech recognition systems must deal with noises and acoustic reverberations in real environments. A microphone array signal processing and model adaptation method are proposed for distant-talking speech recognition in noisy reverberant environments. However, speech sounds captured by a microphone array are distorted by the directional gain patterns of the microphone array and reverberations in the room. Furthermore, model adaptation would give better performance with high SNR. This paper proposes a method to combine microphone array signal processing with model adaptation methods. A speech recognition expenment in a real room showed that the proposed method provides better performance than conventional methods.

Full Paper

Bibliographic reference.  Nishiura, Takanobu / Miki, Kazuhiro / Nakamura, Satoshi / Shikano, Kiyohiro (2001): "Complimentary combination of microphone array and HMM composition for noisy speech recognition", In HSC2001, 167-170.