A Comparison of a Hardware and a SoftwareIntegrate and Fire Neural Network

Abstract

Onset clustering (which we use as part of a system for sound seg-mentation) uses integrate-and-re neurons to perform across spectrum and across time clustering of increases in sound intensity in diierent parts of the spectrum. We show that a network of recently developed analogue VLSI integrate-and-re neurons can perform this task in real-time, and compare its performance with a simulated network. 1 Background A sound wave is a pressure wave, that is a variation in air pressure over time. Virtually all attempts at interpreting sound start by separating the sound out into its constituent frequencies. The mammalian ear 4] is no exception to this, and the cochlea in the inner ear of mammals performs a mechanical ltering of the sound, resulting in a pattern of vibrations on the basilar membrane, a membrane which runs the length of the cochlea. Vibrations at high frequencies are much stronger at the basal end of the cochlea, and low frequencies at the apical end. Transduction of these vibrations into signals on the auditory nerve is performed by the inner hair cells of organ of Corti (which stretches along the length of the cochlea) and the neurons of the To whom correspondence should be sent

Cite this paper

@inproceedings{Smith2007ACO, title={A Comparison of a Hardware and a SoftwareIntegrate and Fire Neural Network}, author={Leslie S. Smith and Mark A. Glover and Alister Hamilton}, year={2007} }