T.J.Moir Personal pages
Independent Component Analysis
(C)2011 by T.J.Moir. All rights reserved.
ICA has been in existence since around the mid 1990s and
is a method to separate (in this case) two or more acoustic
sources which have become mixed together eg whilst recording or
in speech recognition problems where one of the sources is unknown
non-stationary noise or a disturbance. The assumption is made that
the original sources to be recovered are statistically independent.
Moreover the scaling of the original sources cannot be recovered.
The method is often called unsupervised adaptive filtering in the signal processing literature.
The method that concerns us is the Natural Gradient method for convolutive mixtures. The film
below illustrates the approach in the time-domain. Note that faster frequency-domain approaches are also
available in the literature. The pictures illustrates the basic problem with room reverberation.
Essentially the de-mixing (deconvolution) filters needed to un-mix such realistic sound mixtures
can be quite large (maybe 1000 weight matrices).
A film illustrating the Natural Gradient Method for Convolutive Mixtures
Mail me at .... :tom@speechresearch.co.nz