LOCALIZATION BASED SEPARATION OF MOVING AUDIO SOURCES FROM STEREO MIXTURES BY ADAPTIVE BEAMFORMING
Khademul Islam Molla, Keikichi Hirose, Nobuaki Minematsu · 2006
The separation of moving audio sources from their mixtures is practically more useful as well as challenging in the area of source separation. In this paper, a microphone pair is used to capture the signals of multiple moving audio sources. It is assumed that the sources are static within a small time slot. Then localization based separation from the mixtures is obtained by applying adaptive beamforming within that slot. A coherence based source localization method is proposed here. When the sources are moving randomly, it is obvious to arise a permutation problem among the sources separated from the successive time slots. A data- adaptive audio source discrimination method is introduced to resolve such problem. Then the overall separation is obtained by concatenating the sources segregated from consecutive time slots. The experimental results are presented to demonstrate the performance of the proposed method.