Nova Técnica de Separação Cega de Fontes em Subbandas

In this paper we investigate the use of subband filte- ring in blind source separation (BSS) for algorithms convolutive mixtures. The proposed method is based on the frequency domain implementation of the TRINICON algorithm, which combines the advantages of wideband algorithms with the computational efficiency of narrow band techniques. The required modifications in the updating equations of the separation system coefficients are presented to perform them using the signals of the mixtures decomposed into subbands, providing an improvement in the con- vergence rate, higher signal-interference ratio at steady state and lower computational cost in relation to the fullband algorithm. Simulation results demonstrate the advantages of the proposed method in the separation of voice signals in environments with high reverberation times.