---------------------------------------------------------
*** Contact Info:
Maximo Cobos
Department of Computer Science
University of Valencia, Spain.
webpage: http://www.uv.es/macose2
e-mail: Maximo.Cobos@uv.es

---------------------------------------------------------------------
*** Basic information about the algorithm:
Unfortunately, the algorithm is not published yet. At this moment, I am 
working on the manuscript describing the algorithm, with the following 
temptative title:

M. Cobos, "Separation of Instantaneous Underdetermined Speech Mixtures 
in a Transformed Spatial Domain", in preparation, 2011.

Brief Description:

1 - Mixing matrix estimation based on amplitude-ratio peak picking.
2 - A monte-carlo model for the mixture is constructed using a set of 
dependent sparse distributions.
3 - An optimum spatial transformation of the mixture is selected by 
looking at the resulting overlap of amplitude-ratio distributions 
computed from the monte-carlo model.
4 - The sources are extracted using conventional time-frequency masking 
(DUET-like approach) performed on the transformed spatial domain.

----------------------------------------------------------------------
***  Compuation time:
approximately 7.14 s/(test*GHz)  for  2-channel mixtures (test2, dev2)
 approximately 2357 s/(test*Ghz) = 39.2 min/(test*Ghz)  for 3-channel 
mixtures (test3,dev3)