A Single Channel Subspace Speech Enhancement Approach Based on Optimal Lagrange Multiplier in Time Domain Constraint

Abstract

A single channel subspace speech enhancement approach based on optimal Lagrange multiplier in time domain constraint is proposed. The proposed method focuses on the time domain constraint. The inverse of the Cholesky factorization matrix of noise covariance matrix is used to prewhiten the noisy signal. In this paper, the noise suppression level is adjusted adaptively and is decreasing monotonically according to a short time signal to noise (SNR), then the optimal Lagrange multiplier is obtained by a numerical algorithm with high speed. Simulation shows that the proposed approach outperform conventional subspace methods providing high noise reduction and better speech quality in most cases.