Speech filtering for improving intelligibility in noisy transients

Abstract

Hearing impairment is a problem that affects a large percentage of the population. Cochlear implants allow those with profound or total hearing loss to regain some hearing by stimulating auditory nerve fibers with implanted electrodes, in response to sound picked up by an external microphone. The signal processing chain from microphone input to stimulation output is an important factor in the overall speech intelligibility of the implant system. This thesis work improves on an existing ultra-low-power cochlear implant system by utilizing an improved noise and power efficient bandpass filter bank to implement a novel frequency-selective gain control algorithm capable of reducing, and in some cases removing, loud transient noises, thereby improving speech intelligibility. This gain control algorithm takes advantage of the inherent frequency-specific gain control afforded by the improved bandpass filter topology. This contribution makes an improvement to the existing state-of-the-art system in both power efficiency and performance.