Co-Investigator(Kenkyū-buntansha) |
LEE Akinobu Nara Institute of Science and Technology, Graduate School of Information Science, Assistant Professor, 情報科学研究科, 助手 (80332766)
SHIKANO Kiyohiro Nara Institute of Science and Technology, Graduate School of Information Science, Professor, 情報科学研究科, 教授 (00263426)
|
Research Abstract |
In recent years, an accuracy in the speech recognition system can be remarkably improved by using Hidden Markov Model and Neural Networks. However, in real environments, there still exists the significant problems that the speech recognition performance degrades because of the additive noise and reverberation of the room. In this study, we introduce a microphone array technology in which the sound sources can be identified and detected accurately in the three-dimensional acoustic field. This study can provide the following final results. (1) The real acoustic database has been constructed using a 56-ch microphone array system, and the database has been widely distributed. (2) The accurate DOA (direction of arrival) estimation technique with the CSP method has been realized. In addition, we have applied the technique into a moving robot navigation problem. 3) As a new array signal processing, we have proposed the multi-beamforming technique and the blind source separation by ICA. The effectiveness has been revealed through the experiments in the real situations. (4) We have proposed a new approach that combines three-dimensional Vitabi search in the speech recognition and DOA estimation. The research results have been published as follows ; Journal paper 12, International conference 30, Invited talk4, Technical Report 11, Domestic workshop 15.
|