A Noise Type Classifier for Speech Enhancement in DCT Domain

Abstract

Traditional speech enhancement algorithms are mostly attenuating filters, which introduce speech distortion when dealing with destructive noise. Some scholars have proposed a double-gain filter for speech enhancement, which considers the two types of noise coefficients, namely constructive noise and destructive noise. Since it is a double-gain filter, the type of noise needs to be classified before using the corresponding filter. Based on the difference of mean square error in double-gain filter, a new noise type classifier is proposed in this paper, which can effectively classify the types of noise before using the double-gain function. The simulation results show that the proposed error double-gain filter has improved speech intelligibility and comprehensive speech quality compared with the existing algorithms.