Simon
Arberet
EPFL,switzerland
simon.arberet@epfl.ch
S. Arberet, A. Ozerov and N. Q. K. Duong

ALGORITHM:

1. 30 iterations of  algorithm [1] with the same setting as in [1].
2. 100 iterations of  algorithm [2] with the following particular models were used:

   a. for speech sources: 10 NMF components per source

   b. for music sources:  4 NMF component per source.


INITIALIZATION of algorithm [2]:
   
An initial NMF source decomposition is computed from the variances obtained by [1].
The initials covariance’s matrices of [2] are the one obtained by [1].

COMPUTATIONAL TIME

Our Matlab implementation on Mac Pro quad core intel xeon 2.8GHz, 16GB RAM runs up to 35 minutes for algo [1] and 15 minutes for algo [2].

REFERENCES: 

[1] N. Q. K. Duong, E. Vincent and R. Gribonval, Under-determined convolutive blind source separation using spatial covariance models, Proc. IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Dallas-USA, Mar. 2010.

[2] S Arberet, A. Ozerov, N. Q. K Duong, E. Vincent, R Gribonval, F. Bimbot and P Vandergheynst, Nonnegative matrix factorization and spatial covariance model for under-determined reverberant audio source separation, Proc. International Conference on Information Science, Signal Processing and their Applications (ISSPA. IEEE), Kuala Lumpur, May. 2010, to appear.