Speech Recognition in Additive Noise and Room Acoustics Distortion by HMM Composition

• TAKIGUCHI Tetsuya

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

• NAKAMURA Satoshi

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

• SHIKANO Kiyohiro

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

Bibliographic Information

Other Title

• 加法性雑音、伝達特性による歪みを受けた音声のHMM合成による認識

Abstract

All kinds of noises cause a degradation of speech recognition rates in a real environment. When a speaker utters from distance, the speech would be suffered from room acoustics distortion. The room acoustics distortion also causes a degradation of speech recognition rates. To compensate this degradation, the HMM composition method is adapted to not only an additive noise, but also the room acoustics distortion. A HMM attempts to model the room acoustics distortion. A state of the room acoustics distortion HMM corresponds to a position of sound sources. This HMM can estimate the position of sound sources, even the speaker moves. Further this paper proposed a method which combines a speech HMM, a noise HMM and a room acoustics distortion HMM to recognize the noisy distorted speech. Results of 500 words recognition experiments were 23.6%, 78.6% and 80.6%, by the clean speech HMM, the speech and noise HMM, and the speech, noise and room acoustics distortion HMM, respectively. The improvement of 57.0% clarified the effectiveness of the proposed method.

Journal

• IEICE technical report. Speech

  IEICE technical report. Speech 95 (319), 41-46, 1995-10-20

  The Institute of Electronics, Information and Communication Engineers

Keywords

• background noises
• room acoustics distortion
• HMM composition

Details

• CRID: 1573387452154383232
• NII Article ID: 110003296614
• NII Book ID: AN10013221
• Text Lang: ja
• Data Source:
  • CiNii Articles