Noise Cancellation Based on Split Spectra by Using Sound Location
Kazuyuki Nobu, Takeshi Koya, Kei-ichi Kaneda, Naomi Haratani and Hiromu Gotanda
Graduate School of Advanced Technology, Kinki University, 11-6, Kayanomori, lizuka-shi, Fukuoka, Japan
Received:August 18, 2002Accepted:September 4, 2002Published:February 20, 2003
Keywords:robot, automatic speech recognition, noise reduction, speech extraction, independent component analysis, permutation problem
Robots with an automatic speech recognition system (ASR) have been developed to improve its communication with people. The recognition ability of ASR becomes significantly impaired by environmental background noises. This paper studies noise reduction by independent component analysis (ICA) under the situation with two sound sources and two microphones, assuming that the target sound source is relatively close to one microphone and the noise source is relatively close to the other Under this assumption, a rule for judging whether permutation occur or not in ICA is derived by use of nice properties of the split spectrum suggested by Ikeda et al. Based on the rule, a new permutation correction is proposed. The target sound spectrum is extracted in the frequency domain by using the permutation correction, and then restored in the time domain by the inverse discrete Fourier transform. From several experiments, it has been confirmed that our proposed method is valid and articulation of the restored speech is good.
Cite this article as:K. Nobu, T. Koya, K. Kaneda, N. Haratani, and H. Gotanda, “Noise Cancellation Based on Split Spectra by Using Sound Location,” J. Robot. Mechatron., Vol.15 No.1, pp. 15-23, 2003.Data files:
Copyright© 2003 by Fuji Technology Press Ltd. and Japan Society of Mechanical Engineers. All right reserved.