Zbynek Koldovsky (1), Petr Tichavsky (2), and Jiri Malek (1)

(1)
Faculty of Mechatronic and Interdisciplinary Studies
Technical University of Liberec, Studentska 2, 461 17 Liberec,
Czech Republic
{jiri . malek, zbynek . koldovsky} (at) tul . cz

(2)
Institute of Information Theory and Automation,
Pod vodarenskou vezi 4,
P.O.Box 18, 182 08 Praha 8, Czech Republic
tichavsk (at) utia . cas . cz

The separation was performed by the T-ABCD algorithm from [1,2] using Laguerre 
filters with mu=0.7. The BGSEP algorithm was applied in the ICA stage of the 
method. Only first 8000 samples (0.5 second) of each input signals were used 
for the computations of separating filters. The other parameters were L=20 (the 
length ofseparating filters) and alpha=2 (the weighting parameter). The method 
is a slightly modified version of the original one by using GCC-PHAT coefficients
as a criterion for the similarity of independent components introduced in [3].

The processing is linear, therefore, Task 1 is considered only.

The method was running in Matlab.

Average running time per test: 0.69s
Total running time per GHz of CPU: 65.7s


[1] Zbynek Koldovsky, Petr Tichavsky, and Jiri Malek: Time-domain 
Blind Audio Source Separation Method Producing Long Separating 
Filters, submitted to LVA/ICA 2010, 2010.

[2] Z. Koldovskı and P. Tichavskı, "Time-Domain Blind Separation 
of Audio Sources on the basis of a Complete ICA Decomposition of 
an Observation Space", accepted for publication in IEEE Trans. on Audio, 
Speech and Language Processing, April 2010.

[3] Jiri Malek, Zbynek Koldovsky and Petr Tichavsky, Adaptive 
Time-Domain Blind Separation of Speech Signals, submitted 
to LVA/ICA 2010, 2010.