Evaluation of a Speech Enhancement Strategy with Normal-Hearing and Hearing-Impaired Listeners

Abstract

An adaptive digital signal processing procedure was applied to various speech signals in three noise backgrounds. The procedure uses a modified approach to Wiener filtering to estimate a noise suppression filter which reduces the signal at frequencies most likely to be corrupted by noise. Speech intelligibility was compared with and without processing, for speech presented in three types of background noise: multitalker babble, wide-band noise, and narrow-band noise. The speech signals used were six spondaic words spoken by a male talker; 20 consonants, in an intervocalic environment, spoken by a female talker; and a passage of continuous discourse read by a female talker. Adaptive speech reception threshold (ASRT) testing was used to estimate the minimum SNR at which the spondaic words could be identified 70.7% of the time. With this test, signal processing improved performance by 11 dB in narrow-band noise for normal-hearing listeners; no statistically significant improvement was observed when listening in a background of wide-band or speech babble noise. Five of the six hearing-impaired listeners also had improved performance when listening in narrow-band noise, but minimal changes in wide-band or speech babble noise. With the consonant targets, a closed-set nonsense syllable testing procedure demonstrated that processing did not change or slightly reduced performance for all listeners. With continuous discourse all listeners demonstrated a strong preference for processed signals under most listening conditions. The results indicate that a digital adaptive filter implemented within a preconditioning circuit in a hearing aid may provide benefit for hearing-impaired listeners and that it is likely to be preferred over unprocessed input in many listening conditions.