Estimating Source Dominated Microphone Clusters in Ad-Hoc Microphone Arrays by Fuzzy Clustering in the Feature Space

Abstract:
Microphones in an ad-hoc microphone array provide distributed signal acquisition in an acoustic environment in which multiple sound sources may be active. For some applications, e.g., in the context of signal classification or noise reduction it is helpful if clusters of microphones can be determined which are dominated by one of the sources, and one additional cluster of microphones which receives contributions from all sources with similar power and a high amount of reverberation. We propose to perform a fuzzy clustering of audio features which are extracted from microphone signals and evaluate the fuzzy membership values to assign microphones either to source dominated clusters or to a background cluster. Simulations and recordings of scenarios with two sound sources and multiple microphones are used to evaluate the proposed assignment procedure.