Blind Separation of Acoustic Sources using the Teager-Kaiser Energy Operator

Abstract:
This paper presents a localization based approach for blind source separation. We use a small microphone array and sample the received signals in a time interleaved, round robin fashion. The localization process is based on the Teager-Kaiser Energy Operator to estimate the instantaneous frequency of these circularly sampled signals whose phase depends on the direction of impinging sound waves. By this means we preserve the principle characteristics of the circular array geometry of having no preferred look direction, but process only two circularly sampled signals. Based on such localization results and the corresponding phase patterns in each frequency bin, we present in this work a method to calculate binary weighting matrices in time-frequency domain for blind source separation. Experimental results show that two sources can be successfully separated. However, as the Teager Kaiser operator is very sensitive to noise, more work needs to be done before this approach can be used in adverse environments.