Active source selection using gap statistics for underdetermined blind source separation

We address the problem of automatically determining the number of active sources in underdetermined blind source separation (BSS). A time-frequency approach to underdetermined BSS is exploited to discriminate the time-frequency structure of the measured mixtures. To determine the number of active sources over an observation interval, an advanced clustering technique based on gap statistics is proposed. Simulation studies are presented to support the proposed approach.