Auditory information processing with nerve-action potentials

Neural sound processing requires compression of the huge dynamic range of acoustic signals to the limited dynamic range which can be coded by discrete action potentials. We present an inner ear model followed by a compression stage and a realistic sensory cell model. Frequency analysis was achieved using a wave-digital filter model of the inner ear's hydrodynamics. The dynamic compression stage entailed nonlinear and reasonably broad response curves. The model coded speech signals into trains of nerve-action potentials, which were fed into a network of neurons featuring spectral receptive fields with lateral inhibition. When their receptive field width was changed, these neurons coded spectral aspects of sounds and they detected the formants in speech signals.