[Title]

Frequency-Domain Blind Source Separation Systems for Convolutive Mixtures

[Presenter]

Hiroshi Sawada, NTT Communication Science Laboratories

[Abstract]

Two signal separation systems developed by the presenter were evaluated
in SiSEC 2010.  Both systems are based on the frequency-domain
approach where observed mixtures are converted into frequency-domain
time-series signals.  The first system is for underdetermined
convolutive mixtures [1,2].  The separation is performed by time-frequency
(T-F) masking.  T-F masks are designed by frequency-bin wise
observation vector clustering followed by permutation alignment based
on the posterior probabilities estimated in the vector clustering.
The second system is for determined convolutive mixtures, and employs
frequency-bin wise ICA to separate the mixtures [3,4].  Permutation
alignment is also needed in the second system, and conducted by
considering the power ratio of the separated signals.  Both separation
systems attained good results in a variety of the SiSEC 2010 tasks.

[References]

[1] H. Sawada, S. Araki, S. Makino, "A Two-Stage Frequency-Domain
Blind Source Separation Method for Underdetermined Convolutive
Mixtures," IEEE Workshop on Applications of Signal Processing to Audio
and Acoustics (WASPAA 2007), pp. 139-142, Oct. 2007.

[2] H. Sawada, S. Araki, S. Makino, "Underdetermined Convolutive Blind
Source Separation via Frequency Bin-wise Clustering and Permutation
Alignment," IEEE Trans. Audio, Speech, and Language Processing (in
press) Digital Object Identifier: 10.1109/TASL.2010.2051355.

[3] H. Sawada, S. Araki, S. Makino, "Frequency-Domain Blind Source
Separation," in Blind Speech Separation, S. Makino, Te-Won Lee, and
H. Sawada, Eds, Springer, Sep. 2007.

[4] H. Sawada, S. Araki, S. Makino, "Measuring Dependence of Bin-wise
Separated Signals for Permutation Alignment in Frequency-domain BSS,"
IEEE International Symposium on Circuits and Systems (ISCAS 2007),
pp. 3247-3250, May 2007.