Spectral resolution and speech recognition in noise by cochlear implant users.

Abstract

For cochlear implant (CI) users, the relationship between spectral resolution and speech perception in noise has remained ambiguous. An even more fundamental question has been how to measure spectral resolution in CI listeners. This dissertation describes work exploring the relationships among different measures of spectral resolution, and between each of those measures and speech recognition in quiet and in noise. Spectral ripple discrimination was found to correlate strongly with spatial tuning curves, when the measures were matched in frequency region. Broadband spectral ripple discrimination correlated well with sentence recognition in quiet, but not in background noise. Spectral ripple detection correlated strongly with speech recognition in quiet, but its validity as a measure of spectral resolution was not empirically supported. Spectral ripple discrimination thresholds were compared to sentence recognition in noise, using spectrally-limited maskers that did not overlap with the entire speech spectrum. Speech reception thresholds were measured in the presence of four low- or high-frequency maskers, all bandpass-filtered from speech-shaped noise, and a broadband masker encompassing most of the speech spectrum. The findings revealed substantial between-subject variability in susceptibility to masking by each of these noises and in spectral release from masking, which cannot be explained simply in terms of energetic masking and does not appear to be strongly related to spectral resolution. Better CI users appeared to show stronger relationships between spectral resolution and speech perception than did poorer users, implying that advanced CI processing strategies designed to maximize the number of spectral channels may not benefit all CI users equally.