Authors: Jason Mikiel-Hunter¹, Richard Lyon², Rob Schonberger², Honglin Yu²
¹Macquarie University
²Google Australia
Background: The CARFAC (Cascade of Asymmetric Resonators with Fast-Acting Compression) model simulates how the human inner ear processes sound, specifically mimicking the basilar membrane with digital filters. Although CARFAC v2 effectively reproduced many real-life auditory system observations, its Neural Activity Pattern (NAP) output was equivalent to the average firing rate of a single population of auditory nerve (AN) fibers in each frequency channel. In CARFAC v3, we aim to introduce low-, medium- and high-spontaneous rate auditory nerve fiber classes, offering varied responses to sounds of different intensities.
Methods: To simulate distinct NAPs for each class of AN fibers, we adapted CARFAC v2’s two-capacitor inner hair cell (IHC) model. Separate sigmoidal transfer functions—varying mainly in offset—were implemented for the three AN fiber classes so that neurotransmitter release responses (after the second-capacitor loop) differed for the same IHC receptor potential (voltage at the first-capacitor loop).
Results: We demonstrate the continued “good fit” of the CARFAC v3’s NAP output to the physiological data. Furthermore, using a neurometric model of auditory perception that compares neurogram outputs in response to silence and pure-tones, we identify which cochlear parameters are key to achieve “normal-hearing” audiogram thresholds when a physiologically realistic “middle-ear filter” is applied prior to our inner ear model.
Conclusions: We discuss how a selective, parametrized synaptopathy (reduction in AN fiber number, especially in the low- and medium-spontaneous rate classes) in the CARFAC v3 as a hearing impairment, either by itself (i.e. “hidden hearing loss”) or in conjunction with outer-hair-cell impairment may form the basis for modelling individual hearing losses.
Figure legend: Schematic of CARFAC v3 inner hair cell and ANF synapse stages, displaying on-frequency output to a 40dB 4kHz pure tone input for three AN fiber classes: high spontaneous rate (HSR) / medium spontaneous rate (MSR) / low-spontaneous rate (LSR).
