LAUSR

This paper addresses the problem of speech quality enhancement by adaptive two-channel filtering algorithms. Recently, the forward blind source separation structure has been proposed and combined with normalized least-mean-square algorithm (FNLMS). The main drawback of two-channel FNLMS algorithm is its poor performance in steady state regime when the fixed step-sizes values are selected large. However, the slow convergence rate is observed with the small fixed step-size values. In this paper, we propose three new combinations of the basic FNLMS algorithm with simple variable step-sizes approaches, for improving both the steady state values and convergence rate (noted TVSF for Two-channel Variable Step-size Forward). In these modifications, we propose new configuration of two-channel forward structure by three simple and efficient variable step-sizes estimations. To confirm the good performance of three proposed TVSF algorithms compared with the classical fixed-step-size version, we have carried out several simulations in very noisy situations using several criteria.