Semi-blind source separation via sparse representations and online dictionary learning

This work examines a semi-blind source separation problem where the aim is to separate one source, whose local (nominally periodic) structure is partially or approximately known, from another a priori unspecified but structured source, given only a single linear combi-nation of the two sources. We propose a novel separation technique based on local sparse approximations; a key feature of our proce-dure is the online learning of dictionaries (using only the data itself) which sparsely model the a priori unknown source. We demonstrate the performance of our proposed approach via simulation in a styl-ized audio source separation problem. Index Terms — Semi-blind source separation, sparse represen-tations, online dictionary learning 1