Modeling intelligibility of hearing-aid compression circuits

The active filtering effect in the inner ear is disrupted with sensorineural hearing impairment. This causes a loss of frequency selectivity and dynamic range. Compression is often used in hearing-aids in an attempt to re-establish the normal dynamic range of the cochlear response. While some studies show increased speech intelligibility with artificial noise sources for compressive hearing-aids, most show little (< 1 dB versus linear aids) or no advantage in competing speech. In this paper we explore a quantitative model to explain the empirical performance of compressive hearing-aids in competing speech. By combining an accurate cochlear model with a model of higher auditory feature analysis based on spectral-temporal clustering of onsets, we provide an explanation for the failure of hearing-aid compression algorithms to increase intelligibility. Our proposed spectral-temporal intelligibility model suggests that increasing intelligibility for a hearing impaired person in competing speech requires both spectral and temporal suppression.