Hybrid binaural singing voice separation

In this paper, we examine the possibility of harmonic music source separation, with the incorporation of a dummy head. Fusion of multichannel and single-channel blind source separation algorithms with low computational latency has been used for the isolation of harmonic parts of the music combination, recorded in different azimuth positions around the artificial head. Subsequently, we performed objective and subjective tests to find the best algorithm conjunction which improves the separability of music sources.