Savaşkan Bulek | Searcharxiv

arXiv2017

Multichannel Linear Prediction for Blind Reverberant Audio Source Separation

A class of methods based on multichannel linear prediction (MCLP) can achieve effective blind dereverberation of a source, when the source is observed with a microphone array. We propose an inventive use of MCLP as a pre-processing step for blind source separation with a microphone array. We show theoretically that, under certain assumptions, such pre-processing reduces the original blind reverberant source separation problem to a non-reverberant one, which in turn can be effectively tackled using existing methods. We demonstrate our claims using real recordings obtained with an eight-microphone circular array in reverberant environments.

cs.SD↗

arXiv2016

A Penalty Function Promoting Sparsity Within and Across Groups

We introduce a new weakly-convex penalty function for signals with a group behavior. The penalty promotes signals with a few number of active groups, where within each group, only a few high magnitude coefficients are active. We derive the threshold function associated with the proposed penalty and study its properties. We discuss how the proposed penalty/threshold function can be useful for signals with isolated non-zeros, such as audio with isolated harmonics along the frequency axis, or reflection functions in exploration seismology where the non-zeros occur on the boundaries of subsoil layers. We demonstrate the use of the proposed penalty/threshold functions in a convex denoising and a non-convex deconvolution formulation. We provide convergent algorithms for both formulations and compare the performance with state-of-the-art methods.

math.NA↗