Multi-inconsecutive-frames moving average for the frequency-domain blind source separation of convolutive mixtures

In frequency-domain blind source separation of convolutive mixtures, the DFT coefficients of mixture frame are employed to estimate the mixing matrix on each frequency bin. Based on the relation between partial convolution and circular convolution, we propose a multi-inconsecutive-frames moving average method to produce new frame for DFT to substitute the mixture frame. Its DFT coefficients could estimate the mixing matrix on each frequency bin more precisely than traditional way, which improves the separation performance. Simulation results verify the validity of proposed multi-inconsecutive-frames moving average method for frequency domain CMBSS.