Localization and Source Separation Using Expectation-Maximization
An arbitrary array of microphones is utilized for the joint task of localizing and separating multiple concurrent sources.
The recently presented expectation-maximization (EM) is addressing the first task, namely detection and localization.
Here we extend this algorithm to address the second task, namely blindly separating the sources.
We show that the proposed algorithm is capable of separating sources in reverberation enclosure without a priori information on their number and locations.
In the first stage of the proposed algorithm is applied for blindly detecting the active sources and to estimate their directions.
Separation is obtained by utilizing two by products of the localization. First, the hidden variables of the EM algorithm are used to construct masks for each source.
Second, the minimum variance distortionless response (MVDR) filters are used to reduce the noise.
* This research was carried out towards the PHD. Degree in Electrical Engineering at Bar-Ilan University, under the supervision of Prof. Sharon Gannot.