Blind source separation for convolutive mixtures using spatially resampled observations

We propose a new technique for separation of sources from convolutive mixtures based on independent component analysis (ICA). The method allows coherent processing of all frequencies, in contrast to the traditional treatment of individual frequency bands. The use of an array enables resampling of the signals in such a way that all frequency bands are effectively transformed onto the centre frequency. Subsequent separation is performed “all-bands-in-one”. After resampling, a single matrix describes the mixture, allowing use of standard ICA algorithms for source separation. The technique is applied to the cocktail-party problem to obtain an initial estimate of the separating parameters, which may further be processed using crosstalk removal or filtering. Experiments with two sources of speech and a four element microphone array show that the mixing matrix found by ICA is close to the theoretically predicted, and that 15 dB separation of the sources is achieved.