Determined BSS based on time-frequency masking and its application to harmonic vector analysis
share
When the number of microphones is equal to that of the source signals (the determined situation), audio blind source separation (BSS) is usually performed by multichannel linear filtering to deal with the convolutive mixing process. By formulating the determined BSS problem based on the statistical independence, several methods have been successfully developed. The key to development is the modeling of the source signals, e.g., independent vector analysis (IVA) considers co-occurrence among the frequency components in each source. In this paper, we propose the determined BSS method termed harmonic vector analysis (HVA) by modeling the harmonic structure of audio signals via the sparsity of cepstrum. To handle HVA, the general algorithmic framework that recasts the modeling problem of determined BSS into a design problem of a time-frequency mask is also proposed. Through the experimental investigation, it is shown that HVA outperforms IVA and independent low-rank matrix analysis (ILRMA) for both speech and music signals.
READ FULL TEXT