In this paper, we obtain the ordered speech features by frequency using independent component analysis to human speeches. When independent component analysis is applied to speech signals, the adapted basis functions resemble Gabor-like features.
But in ordinary independent component analysis, the components are assumed to be completely independent, and they do not necessarily have any meaningful order relationships. So this paper introduce ICA model that has a coupling at energy of near components. This coupling could be used to create a ordered speech features in frequency. This features have a similar characteristic with filter banks of basilar membrane except the log-scale distribution.
