Abstract: An algorithm for learning an overcomplete dictionary
using a Cauchy mixture model for sparse decomposition of an underdetermined
mixing system is introduced. The mixture density
function is derived from a ratio sample of the observed mixture
signals where 1) there are at least two but not necessarily more
mixture signals observed, 2) the source signals are statistically
independent and 3) the sources are sparse. The basis vectors of the
dictionary are learned via the optimization of the location parameters
of the Cauchy mixture components, which is shown to be more
accurate and robust than the conventional data mining methods
usually employed for this task. Using a well known sparse
decomposition algorithm, we extract three speech signals from two
mixtures based on the estimated dictionary. Further tests with
additive Gaussian noise are used to demonstrate the proposed
algorithm-s robustness to outliers.
Abstract: A frequency grouping approach for multi-channel
instantaneous blind source separation (I-BSS) of convolutive
mixtures is proposed for a lower net residual inter-symbol
interference (ISI) and inter-channel interference (ICI) than the
conventional short-time Fourier transform (STFT) approach. Starting
in the time domain, STFTs are taken with overlapping windows to
convert the convolutive mixing problem into frequency domain
instantaneous mixing. Mixture samples at the same frequency but
from different STFT windows are grouped together forming unique
frequency groups.
The individual frequency group vectors are input to the I-BSS
algorithm of choice, from which the output samples are dispersed
back to their respective STFT windows. After applying the inverse
STFT, the resulting time domain signals are used to construct the
complete source estimates via the weighted overlap-add method
(WOLA). The proposed algorithm is tested for source deconvolution
given two mixtures, and simulated along with the STFT approach to
illustrate its superiority for fairly motionless sources.