A neural network approach to auditory streaming

Abstract

This paper suggests a neural network architecture for the streaming of an acoustic input using fundamental grouping cues of the mammalian auditory system. The auditory periphery is simulated using biologi- cally relevant models, and the primary auditory cortex is represented by a neural network layer, in which three di erent features of the input are mapped and streamed. Finally, an attentional searchlight that is driven by the cortical network neurons, focuses on a selected cortical area representing a listener who focuses their attention on a distinct acoustic input.