A noise reduction method based on linear prediction analysis
Arata Kawamura, Kensaku Fujii, Yoshio Itoh, Yutaka Fukui · Electronics and Communications in Japan (Part III Fundamental Electronic Science) · 2002
Abstract The linear prediction method works in such a way that the difference (prediction residual) between the objective signal and the output generated by making the objective signal a linear combination of the input signals is uncorrelated. Hence, in this linear prediction method, the linear predictor generates zero for the signal without correlation and recombines the voice for that with correlation. If this characteristic is used, recombined voice with suppressed noise can be obtained as the output of the linear predictor for voice with superposed white noise without correlation. Noise suppression can be realized by using this characteristic of the linear predictor. In this paper, a noise suppression method based on this linear prediction is discussed. According to this method, a noise suppression method without process delay can be realized. Also, no a priori information on the power spectrum of the noise is needed in this method. © 2002 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 86(3): 1–10, 2003; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/ecjc.10056