Content area
Abstract
Listeners with similar degrees of hearing loss often perform quite differently on speech recognition tasks. Why this occurs, however, remains unclear. In this series of studies, how listeners weight spectral information in sentences was assessed. Results are reported for both listeners with normal hearing and listeners with hearing loss.
Harvard/IEEE sentences and noise with a spectrum matched to the sentences were filtered into five spectral bands with approximately equal amounts of intelligibility (based on 1/3-octave band importance functions). On each trial the five respective filtered sentence and noise bands were randomly combined at various signal-to-noise ratios (SNRs). Spectral weights were computed using a point-biserial correlation between a listener's response on the recognition task (correct or incorrect) and the SNR in each frequency band. The stronger the correlation between performance and SNR, the greater that given band contributed to the listeners' performance.
At a conversational speech level, normal-hearing listeners weighted bands 5 and 2 significantly greater than the three other spectral bands. The listeners with sensorineuralhearing loss weighted band 5 significantly greater than all of the other spectral bands. These same listeners were also tested while fit with linear hearing aids. The aided listeners continued to weight the spectral information in band 5 significantly greater than all of the other spectral bands; however, band 5 received significantly less weight in comparison to the unaided condition.
Last, the effect of presentation level on weighting strategies was examined in normal-hearing listeners. Three presentation-levels (55, 75, 96 dB SPL) were tested. As overall-presentation level increased so did the relative weight on band 5. Listeners overall percent-correct scores did not significantly differ between the 55 and 75 dB SPL conditions, but were significantly lower in the 95 dB SPL condition.
