Francesco Nesta
Fondazione Bruno Kessler (FBK-IRST), Trento (Italy)

Main parameters:

FFT size=4096
Frame shift=1024
Number of ICA iterations per frequency=20
Step-size=0.1


Sources have been separated with an algorithm based on a Regularized Recursive ICA (RR-ICA)[1] which extends the method in [2]. 
RR-ICA reduces the permutation if there is no aliasing but wrong permutations may occur if there is spatial aliasing. Thus, for the test files with microphone spacing of 6cm and 10cm the permutation of higher frequencies are corrected by means of the TDOA estimated by the SCT [3].
The separation is obtained with a real-time algorithm (with a variable cpu load, depending on the parameters, of about 20-30%). 
The algorithm separates the sources in the sequence processing independently blocks of 1 second of data (thus assuming that only two sources overlap in each block).
The external permutation across the blocks is solved by considering the TDOA estimated by the SCT.


[1] Francesco Nesta, Piergiorgio Svaizer, Maurizio Omologo, "Convolutive BSS of short mixtures by ICA recursively regularized across frequencies", (accepted) in IEEE  Transactions on Audio, Speech and Language Processing.
[2] Francesco Nesta, Piergiorgio Svaizer, Maurizio Omologo, "A BSS method for short utterances by a recursive solution to the permutation problem", (in proceedings), SAM2008
[3] Francesco Nesta, Piergiorgio Svaizer, Maurizio Omologo, "A novel robust solution to the permutation problem based on a joint multiple TDOA estimation", (in proceedings), IWAENC 2008