Classifying auditory nerve patterns with neural nets: a modeling study with low level signals

Abstract In man, 30,000 fibers of the primary auditory nerve connect the receptor cells of the inner ear with the central auditory nervous system. The acoustic information in the auditory nerve is binary coded: in every fiber up to 400 impulses (spikes) per second are propagated. However, the pattern is disturbed by the spontaneous activity of the nervous system, i.e. without any acoustic signal the high sensitive fibers transfer up to 160 spikes/s. This spontaneous activity seems to be of high importance for detecting low level acoustic signals. The purpose of this study is to use artificial neural network techniques in order to detect any low level auditory information that is hidden in a simulated spiking pattern of the auditory nerve. Sinusoidal stimuli with a signal to noise ratio as low as 1 10 can be recognized from the simulated firing pattern of a single auditory nerve fiber.