A New Frequency Domain Blind Separation and Localization of Anechoic Mixtures Incorporating Spatial Notch Filtering

In this paper blind source separation (BSS)and localization of audio signals from their anechoic mixtures is implemented in frequency domain utilizing the recently developed spatial notch filtering (SNF). The SNF approach which was originally proposed for localization of brain signal sources is combined with complex value frequency domain BSS algorithm. By combining BSS and SNF, localization of every sources and alignment of permutation ambiguity in each frequency bin can be simply performed. By using this approach, the results of the source separation obtained from BSS method is mutually used indescribing the location for each of the source(s) of interest simultaneously. Unlike many other methods for solving the permutation problem meant for speech signals (due to continuity of spectrum for adjacent frequency bins)the method here can be applied to all types of audio sources. Finally, simulation results suggest a significant improvement over the results of previously developed techniques using combined BSS and beam forming method and the comparison results was illustrated.