Speech enhancement based on 4th order cumulant backward linear predictor for impulsive noise
Naoto Sasaoka, Kazumasa Ono, Yoshio Itoh · 2012
Although a speech enhancement has been proposed, it is difficult to reduce the impulsive noise whose power changes fast. In this paper the speech enhancement based on a linear predictor with 4th order cumulant adaptive algorithm is proposed to reduce the impulsive noise which is suddenly generated by hitting an object. The proposed method takes advantages of the high kurtosis of the impulsive noise. The proposed linear predictor converges such that the output of the linear predictor becomes the component with high kurtosis. The impulsive noise has high kurtosis, on the other hand speech has low kurtosis. Therefore, the proposed linear predictor can estimate only the impulsive noise. In addition, in order to improve a convergence rate the backward linear predictor is used herein.