Sue Becker and Ian C Bruce (ed.) (2002)
Neural coding in the auditory periphery: Insights from physiology and modelling lead to a novel hearing compensation algorithm
Neural Information Coding (NICE) workshop, Les Houches, France.
Moderate to severe degrees of noise-induced or age-related hearing loss are associated with impairment of hair cells in the cochlea, while profound hearing loss often involves complete degeneration of the hair cells and of the afferent auditory nerve (AN) fibers that innervate the cochlea. The inner hair cells (IHCs) are involved in the transduction of travelling waves on the basilar membrane into neuronal firing of frequency-tuned AN fibers. In the absence of outer hair cell (OHC) modulation, the IHCs' frequency tuning curves area fairly broad and linear. Both the OHCs and the IHCs, along with the synapses from the IHCs to the AN fibers, introduce nonlinearities in the coding and transmission of several stimulus properties including loudness, frequency and phase. OHCs are responsible for loudness compression, sharpening the frequency tuning curves of inner hair cells in the presence of noise, and other masking and contrast enhancement effects such as synchrony capture at moderate stimulus levels. The IHC and synaptic nonlinearities contribute to synchrony capture at high stimulus levels. Ian Bruce, Laurel Carney and colleagues have developed a phenomenological model that captures these critical nonlinearities in the early coding of auditory signals, based on electrophysiological data recorded from cat auditory nerve fibers, both in the normal and damaged cochlea. We describe this model, and its application to the design of a novel hearing compensation algorithm. The central idea of the compensation algorithm is to design a pre-processor for the damaged model that causes its output to more closely resemble that of the intact model. Traditional hearing aids simply amplify signals on a frequency-by-frequency basis according to measured thresholds for noise-free pure tones. For noisy signals (e.g. the "Cocktail party problem") this strategy may actually worsen intelligibility. Our model-based method should be able to predict and compensate for masking effects due to loss of hair-cell nonlinearities.