Feature extracting hearing aids for the profoundly deaf using a neural network implemented on a TMS320C51 digital signal processor
J. R. Walliker, J. Daley, Andrew T. Faulkner, Ian Spencer Howard · 1991
Many people with profound hearing impairment, while able to detect amplified sound are often unable to make sense of what they hear.Conventional hearing aids which amplify, filter and compress the speech signal are of little use to them.It has been demonstrated that some profoundly deaf listeners are able to make better use of speech features such as voice fundamental frequency (Fx) and frication when they are presented in a simplified form matched to their residual hearing than when conventionally presented.An algorithm is required for this purpose which can extract Fx in real-time, on hattery powered portable hardware.The method should preferably operate on a period-by-period basis both to minimise the processing delay and because there may be useful information in the period-by-period irregularities of a speakers Fx.For the technique to be useful, the extraction method must perform better in a noisy and reverberant environment than the conventionally aided listener.