Research on Speech Signal Enhancement Technology Based on Progressive Deep Neural Network

Abstract

Traditional speech enhancement methods have some shortcomings when processing sound signals. For example, under low signal-to-noise ratio, the residual of music noise is heavy; the effect of non-stationary noise suppression is not obvious; they are all based on the assumption that the noise is a stationary process. In this paper, the progressive speech enhancement method based on neural network, through experimental comparison, can effectively improve the problems of traditional speech enhancement methods.