Speech Enhancement Using a Technique of Adaptive Bias Suppression
Hirokazu Tanaka, Naoya Yamamura, Yuichiroh Ohhashi, Tetsuya Shimamura · 2006
In this paper, we assume speech corrupted by white Gaussian noise and propose a technique of adaptive bias suppression. A linear predictor is used as the basic filter and gamma-NLMS (normalized least mean square) algorithm, which is useful for suppressing bias, is adopted for the predictor adaptation. In addition, for suppressing bias more effectively we apply a weighting parameter and filter banks to the predictor. Experiments on continuous speech result in that the proposed predictor provides superior performances