Dionyssos Kounades-Bastian, Laurent Girin, Xavier Alameda-Pineda, Sharon Gannot and Radu Horaud
[Could not find the bibliography file(s)The separation of moving sound sources is a challenging task mainly because it is extremely complex to devise algorithms that robustly discriminate those signal variations due to the intrinsic variation of the sound source from those signal variations due to the time-varying source-to-microphone channel. We successfully investigated this scenario and got the Best Student Paper Award at IEEE WASPAA’15 [?] and an article at IEEE TASLP to appear in 2016 (whose title/abstract read below) [?].
We proposed a novel probabilistic framework based on the complex Gaussian model combined with non-negative matrix factorization (NMF) for sound source separation. The idea is to model the properties associated with moving sources using time-varying mixing filters described by a stochastic temporal process. We present a variational expectation-maximization (VEM) algorithm that employs a Kalman smoother to estimate the mixing filters. The sound sources are separated by means of Wiener filters, built from the estimators provided by the proposed VEM algorithm. Preliminary experiments with simulated data show that, while for static sources we obtain results comparable with the baseline method, in the case of moving source our method outperforms a piece-wise version of the baseline method.