A novel subspace speech enhancement approach based on test of hypothesis and masking properties

Abstract

A novel signal subspace algorithm based on test of hypothesis and masking properties of the human auditory system is proposed for microphone array speech enhancement. Different from conventional empirism dependent methods, first, for the optimal subspace selection, the paper determines the subspace dimension via solving a test of hypothesis. Then, according to the characters of the speech signal eigenvalues, we use conditional probability to compute the noise variance. At last, the masking properties of human auditory mechanism are incorporated in the subspace approach to estimate the linear filter. Simulation results show the superiority of our proposed novel approach over other existing speech enhancement algorithms in terms of four objective speech quality measures.