Incremental Method of Permutation Alignment For Frequency-Domain Blind Source Separation

This study proposes an incremental permutation-alignment method for frequency-domain (FD) blind source separation (BSS). This method is used to extract the elements belonging to a distinct source at all frequency bins and to reduce the permutation-alignment problem of the N-channel to that of the (N − 1)-channel. Permutation ambiguity is resolved by repeating the extraction. According to the simulation, the proposed method combined with a sparsity-based FD independent component analysis method proposed in 2020 is competitive in terms of the signal-to-distortion ratio with the adoption of recent FD-BSS approaches such as auxiliary function-based independent vector analysis and independent low-rank matrix analysis that can handle inter-frequency dependencies without using permutation alignment.