A Lateral-Inhibitory-Network Model of the Central Processing of Auditory Nerve Activity

Abstract

Lateral inhibitory networks (LINs) of neurons are thought to be prominent in sensory systems and are known to enhance spatial edges and peaks in their input excitation patterns. It is postulated, based on experimental findings, that lateral inhibition contributes to central, sub-cortical, auditory processing. Previous computational LIN models of the central processing of auditory nerve activity were based on highly simplified, non-spiking models of neurons. A more biologically realistic LIN model of spiking neurons was thus developed to investigate the plausibility of such networks achieving contrast enhancement and speech feature extraction. The model developed is a single-layer, uniform, recurrent LIN structure. Each neuron in the LIN is described by a leaky integrate-and-fire model with conductance-based synaptic input. Input spike instances were obtained from Bruce and colleagues' [2003] model of the auditory periphery for synthesized speech stimuli or from a Bernoulli approximation of a Poisson process to represent spontaneous activity from the auditory nerve. The effect of neural and network parameters on contrast enhancement exhibited in the mean spike rates was measured. It was found that the spiking LIN is able to achieve contrast enhancement if the values of the neuronal parameters fall within a very specific and narrow range. Furthermore, the spatial edge in the input had to be high and steep. Compared to non-spiking neuron models, it is quite difficult for spiking neurons to achieve contrast enhancement. The spiking LIN was found to be capable of making formant frequencies more distinct in the average rate profiles of speech stimuli presented at high intensities. However, synchrony of the neural activity to the formant frequencies was largely lost. This architecture of spiking neurons is therefore unlikely to be how contrast enhancement and speech feature extraction is realized in the central auditory system.