5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

An Effect of Adaptive Beamforming on Hands-Free Speech Recognition Based on 3-D Viterbi Search

Takeshi Yamada, Satoshi Nakamura, Kiyohiro Shikano

Graduate School of Information Science, Nara Institute of Science and Technology, Japan

To integrate the microphone array processing into speech recognition, we have proposed a speech recognition algorithm based on 3-D Viterbi search, which localizes a target talker considering the likelihood of HMMs (Hidden Markov Models) while performing speech recognition. The performance of the 3-D Viterbi search method depends on the improvement of the SNR (Signal to Noise Ratio) by the beamforming technique. This paper proposes a novel method based on an adaptive beamforming technique instead of the delay-and-sum beamformer used in our previous study. The speaker-dependent isolated-word recognition experiments were carried out on real environment data to evaluate the effect of the adaptive beamformer. These results showed that the adaptive beamformer drastically improves the recognition performance both for a fixed-position talker and for a moving-talker.

Full Paper

Bibliographic reference.  Yamada, Takeshi / Nakamura, Satoshi / Shikano, Kiyohiro (1998): "An effect of adaptive beamforming on hands-free speech recognition based on 3-d viterbi search", In ICSLP-1998, paper 0484.