Thresholding wavelet-based forward BSS algorithm for speech enhancement and complexity reduction

This paper presents a new speech enhancement method based on wavelet thresholds. Previously, we have proposed the forward blind source separation (FBSS) structure implemented in wavelet transform domain (WFBSS) to reduce acoustic noise components in a two-sensor configuration system. In the simulations results, The WFBSS algorithm has shown superiority in term of convergence speed in comparison with the FBSS algorithm. However, this algorithm has a heavy computational complexity. This problem may be reduced by applying a thresholding method. To achieve this objective, we propose, in this paper, a Two-channel threshold Wavelet-based forward blind source separation algorithm (TWFBSS) algorithm. Performance of the proposed algorithm is evaluated on the estimated speech signal by several criteria in real conditions.