A frequency co-channel adaptive algorithm for speech quality enhancement

摘要

In this paper, we propose a new transform domain implementation of the backward blind source separation (BBSS) structure combined with the normalized least mean square NLMS algorithm. The proposed algorithm uses the short time Fourier transform to get best performances in the frequency domain in comparison with its time domain version. The performance improvements of the proposed frequency-domain BBSS (FD-BBSS) algorithm are low complexity and fast convergence speed properties. In order to see well the performance of the proposed FD-BBSS algorithm, we compare its performances with other competitive algorithms and we use objective criteria to evaluate their behavior in speech enhancement application when very noisy observations are available.