Multichannel sound source dereverberation and separation for arbitrary number of sources based on Bayesian nonparametrics

摘要: Multichannel signal processing using a microphone array provides fundamental functions for coping with multi-source situations, such as sound source localization and separation, that are needed to extract the auditory information each source. Auditory uncertainties about degree of reverberation number sources known degrade performance or limit practical application processing. Such must therefore be overcome realize general robust These uncertainty issues have been partly addressed—existing methods focus on either issue, where joint separation dereverberation has achieved only overdetermined conditions. This paper presents an all-round method achieves arbitrary including underdetermined Our uses Bayesian nonparametrics infinitely extensible modeling flexibility so bypass model selection in problem, which is caused by uncertainty. Evaluation task various numbers conditions demonstrates (1) our applicable mixtures, (2) extraction comparable state-of-the-art suitable