Sound files for the paper

MAXIMUM LIKELIHOOD APPROACH FOR BLIND AUDIO SOURCE SEPARATION
USING TIME-FREQUENCY GAUSSIAN SOURCE MODELS
C. Févotte and J.-F. Cardoso


Submitted to 2005 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics


Noise-free determined mixture (Section 4.1)


Mix 1      Mix 2       Mix 3      Mix 4

JADE      Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
BGML     Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
TFGML   Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4

Noisy determined case (Section 4.2)

Mix 1      Mix 2       Mix 3      Mix 4

Oracle           Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
Oracle diag   Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
TFGML        Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
JADE           Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4

Noisy underdetermined case (Section 4.3)


Mix 1      Mix 2       Mix 3

Oracle           Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
Oracle diag   Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4
TFGML        Estimated source 1     Estimated source 2     Estimated source 3    Estimated source 4





Original source 1       Original source 2       Original source 3       Original source 4