Pdf of Shiro Ikeda's talk:
http://www.mns.brain.riken.go.jp/~shiro/ikeda2001hearnet.pdf

The paper which is deeply related to the talk is

http://www.mns.brain.riken.go.jp/~shiro/papers/etc/blindsep.ps.gz
http://www.mns.brain.riken.go.jp/~shiro/papers/etc/blindsep.pdf
(both are the same except the format.)

ICA results of the speech separations are available from
http://www.mns.brain.riken.go.jp/~shiro/blindsep.html

and if there are any people interested in MATLAB or C codes, I can
send the codes. 

Examples of methods which implement the idea of sparseness of time-frequency domain:

http://www.princeton.edu/~srickard/bss.html

and

http://people.ac.upc.es/pau/