Francesco Nesta 1)Fondazione Bruno Kessler, Trento (TN), Italy 2) Università di Trento (TN), Italy nesta@fbk.eu OVERVIEW The algorithm separates the sources by performing a linear filtering based on ICA. A recursive frequency-domain BSS (see ref. 3-4) is implemented and the permutation are solved by means of a combined TDOA estimation and a recursive spectral estimation. PERFORMANCE For estimating deconvolution filters long 4096 taps (with fs=16kHz): 1) MATLAB (not optimized, too many cycles) 2 Mixtures (2 signals) about 18 s 3 Mixtures (3 signals) about 25 s 4 Mixtures (4 signals) about 32 s 2) C++ (real-time demo for two sources) 2 Mixtures about 3 s CPU: Intel Core 2 T8100 (2.1 Ghz) The number of the sources can be estimated by means of the main peaks of the SCT transform (see ref. 1-2) However in these tests such information has not been explicitely used since we assumed a number of sources equal to the number of the microphones. Main references: 1) Francesco Nesta, Maurizio Omologo, Piergiorgio Svaizer \"Multiple TDOA estimation by using a state coherence transform for solving the permutation problem in frequency-domain BSS\", (in proceedings), MLSP 2008 2) Francesco Nesta, Maurizio Omologo, Piergiorgio Svaizer \"A novel robust solution to the permutation problem based on a joint multiple TDOA estimation\", (in proceedings), IWAENC 2008 3) Francesco Nesta, Piergiorgio Svaizer, Maurizio Omologo, \"Separating short signals in highly reverberant envirnonment by a recursive frequency-domain BSS\", (in proceedings), HSCMA 2008 4) Francesco Nesta, Maurizio Omologo, \"A BSS method for short utterances by a recursive solution to the permutation problem\", (in proceedings), SAM2008