BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Model-based EM Source Separation and Localization in Reverberant M
 ixtures - Michael Mandel ( Columbia University)
DTSTART:20081215T130000Z
DTEND:20081215T143000Z
UID:TALK15644@talks.cam.ac.uk
CONTACT:Rachel Fogg
DESCRIPTION:This talk will describe our system for separating and localizi
 ng multiple sound sources from a reverberant two-channel recording.  The t
 alk begins with a characterization of the interaural spectrogram for singl
 e source recordings\, and a method for constructing a probabilistic model 
 of interaural parameters that can be evaluated at individual spectrogram p
 oints.  Multiple models can then be combined into a mixture model of sourc
 es and delays\, which reduces the multi-source localization problem to a c
 ollection of single source problems.  The talk will then outline an expect
 ation maximization algorithm for finding the maximum-likelihood parameters
  of this mixture model\, and show that these parameters correspond well wi
 th interaural parameters measured in isolation.  As a byproduct of fitting
  this model\, the algorithm creates probabilistic spectrogram masks that c
 an be used for source separation.  In experiments performed in simulated a
 nechoic and reverberant environments\, our system on average improved the 
 direct-path signal-to-noise ratio of target sources by 2.1 dB more than tw
 o comparable algorithms.
LOCATION:LR5\, Engineering\, Department of
END:VEVENT
END:VCALENDAR