Improvement of Speech Perception for Hearing-Impaired Listeners

Abstract

Hearing impairment is becoming a prevalent health problem affecting 5% of world adult populations. Hearing aids and cochlear implant already play an essential role in helping patients over decades, but there are still several open problems that prevent them from providing the maximum benefits. Financial and discomfort reasons lead to only one of four patients choose to use hearing aids; Cochlear implant users always have trouble in understanding speech in a noisy environment. In this dissertation, we addressed the hearing aids limitations by proposing a new hearing aid signal processing system named Open-source Self-fitting Hearing Aids System (OS SF hearing aids). The proposed hearing aids system adopted the state-of-art digital signal processing technologies, combined with accurate hearing assessment and machine learning based self-fitting algorithm to further improve the speech perception and comfort for hearing aids users. Informal testing with hearing-impaired listeners showed that the testing results from the proposed system had less than 10 dB (by average) difference when compared with those results obtained from clinical audiometer. In addition, Sixteen-channel filter banks with adaptive differential microphone array provides up to six-dB SNR improvement in the noisy environment. Machine-learning based self-fitting algorithm provides more suitable hearing aids settings. To maximize cochlear implant users’ speech understanding in noise, the sequential (S) and parallel (P) coding strategies were proposed by integrating high-rate desynchronized pulse trains (DPT) in the continuous interleaved sampling (CIS) strategy. Ten participants with severe hearing loss participated in the two rounds cochlear implants testing. The testing results showed CIS-DPT-S strategy significantly improved (11%) the speech perception in background noise, while the CIS-DPT-P strategy had a significant improvement in both quiet (7%) and noisy (9%) environment.