Detection method of speech signal based on the measurement of body conducted sound for speech recognition under noise environment

2014 Fiscal Year Final Research Report

• Project/Area Number: 24560529
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Measurement engineering
• Research Institution: Prefectural University of Hiroshima
• Principal Investigator: IKUTA Akira 県立広島大学, 経営情報学部, 教授 (30145164)
• Co-Investigator(Kenkyū-buntansha): ORIMOTO Hisako 県立広島大学, 経営情報学部, 講師 (80533207)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 音声信号 / 雑音除去 / 体内伝導音 / 気導音 / 実環境 / 音声認識

Outline of Final Research Achievements

For applying speech recognition to actual circumstances where hand-writing is difficult, a detection method of speech signal was considered under existence of surrounding noise. More specifically, a signal processing method to estimate only the speech signal based on the observation data contaminated by noise was proposed by introducing Beta distribution appropriate for the finite amplitude fluctuation range of signal and observation. Next, both body- and air-conducted speech signals were used simultaneously to fight up against the surrounding noise to improve the enhancement quality. Nonlinear adaptive noise canceller (ANC) was proposed by using FLANN and Volterra filter. Furthermore, a hybrid system combining both FLANN and Volterra filter was proposed. The effectiveness of the proposed system was confirmed by applying it to real body- and air-conducted speech signals measured in an anechoic chamber under existence of surrounding noise.

Free Research Field

音響信号処理