Intertalker differences in intelligibility of cochlear-implant simulated speech

Abstract

Are the acoustic-phonetic factors that promote highly intelligible speech invariant across different listener populations? Two approaches have been taken to investigate intelligibility variation for a variety of listener populations including hearing-impaired listeners, second language learners, and listeners with cochlear implants: studies on how speaking style affects intelligibility and other research on how inherent differences among talkers influence intelligibility. Taking the second approach, we compared intertalker differences in intelligibility for normal-hearing listeners under cochlear implant (CI) simulation (n=120) and in quiet (n=200). Stimuli consisted of 20 native English talkers’ productions of 100 sentences. These recordings were processed to simulate listening with an eight-channel CI. Both clear and CI-processed tokens were presented to listeners in a sentence transcription task. Results showed that the most intelligible talkers in quiet were not the most intelligible talkers under CI simulation. Furthermore, listeners demonstrated perceptual learning with the CI-simulated speech but showed little learning in the quiet. Some of the acoustic-phonetic properties that were correlated with intelligibility differed between the CI-simulated speech and the speech in the quiet. These results suggest that the intertalker variations that result in highly intelligible speech observed in earlier studies are dependent on listener characteristics. [Work supported by NIDCD.]