In real world, a large set of mixed signals are available from which each source signal need to be recovered and this problem can be addressed with adaptive dictionary method. In the case of multichannel observations sparsity found to be very useful for source separation. The problem exist is that in most cases the sources are not sparsified in their domain and it will become necessary to sparsify the source by using some known dictionaries. In order to recover the sources successfully a prior knowledge of the sparse domain is required, if not available this problem can be solved by using dictionary learning technique into source separation. The proposed method, a local dictionary is adaptively learned for each source separately along with separation. This approach improves the quality of source separation both in noiseless and different noisy situations. The advantage of this method is that it denoise the sources during separation.
Independent low-rank matrix analysis based on complex student's t-distribution for blind audio source separation