Structured Sparse Acoustic Modeling for Speech Separation
A novel formulation of acoustic multipath is proposed for estimation of the room acoustic using recordings of unknown concurrent speech sources at unknown locations. The framework exploits sparsity and low-rank structures characterized by the Image method for estimation of the geometry and the absorption factors of the reflective surfaces. The experiments conducted on real data recordings demonstrate the effectiveness of the method for modeling the room acoustic and its application for speech separation and dereverberation.
Record created on 2015-03-19, modified on 2016-08-09