Coherent ICA: Implications for auditory signal processing

In this paper, we describe a novel algorithm, called Coherent independent Components Analysis, and referred to as Coherent ICA for short. The algorithm, rooted in information-theoretic learning, exploits the combined use of the Infomax and Imax principles. Experimental results, based on the auditory coding of natural sounds, are presented that demonstrate the ability of coherent ICA to extract the envelope of amplitude-modulated sounds in a manner similar to the behaviour of neurons in the cochlear nucleus and inferior colliculus.