Masashi Nakayama, Shunsuke Ishimitsu, Seiji Nakagawa
38th International Congress and Exposition on Noise Control Engineering 2009, INTER-NOISE 2009 1 345-349 2009年12月1日
Engineers working in the engine rooms of ships are unable to communicate using normal speech. The engine room becomes a noise environment of engine and generators. To extract a normal speech produced in such noisy environments, we investigated construction of the body-conducted speech recognition system. Body-conducted speech is classified into solid propagated sound on skin or bone; thus it is not influenced by noise in the air. Body-conducted speech produces a low quality signal, because it does not include high frequency components of 2 kHz and greater. For this reason, many researchers have proposed signal retrieval methods, but conventional methods require both normal speech and body-conducted speech. In this paper, we propose the signal retrieval technique that does not require normal speech. First, the method converts the emphasized signal into body-conducted speech as a differential acceleration signal. Second, conventional noise reduction is applied to the emphasized signal to produce the retrieval signal. Our previous research investigated body-conducted speech under calm. The experiment confirmed the effectiveness of the new body-conducted speech retrieval technique, extracting in 98 dB SPL (-20 dB SNR). We conclude that the signal improves in sound quality on frequencies of 2 kHz and greater.