Date on Senior Honors Thesis


Document Type

Senior Honors Thesis

Degree Name



Psychological and Brain Sciences

Degree Program

College of Arts and Sciences

Author's Keywords

cochlear implants; spectral contrast effects; channels; spectrally degraded vowels; noise vocoding; varying talkers


When a person listens to a context sentence with prominent higher frequencies, the subsequent vowel sound is more likely to be perceived as being of a lower frequency and vice versa. This is a spectral contrast effect (SCE). Recent work has shown that talker variability diminishes these SCEs. They were found to be smaller when 200 sentences were spoken by a different talker each time compared to one talker (Assgari & Stilp, 2015). Cochlear Implant (CI) users’ speech categorization is also influenced by SCEs but are known to struggle with talker discrimination. Here, I tested whether talker variability affected SCEs when the speech was spectrally degraded. Listeners indicated whether they heard “ih” (as in bit) or “eh” (as in bet) following 200 context sentences spoken by the same talker or varying talkers (Assgari & Stilp, 2015). The sentences were noise vocoded to broadly simulate CI processing. SCEs occurred but did not significantly differ across one-talker and 200-talker conditions. Talker variability does not appear to affect perception in acoustic simulations of CI processing in the same way it does for normal-hearing listeners.

Lay Summary

Participants sat a computer in a sound booth with headphones, heard a sentence with a subsequent vowel sound and had to indicate which vowel sound they heard. The sentences that preceded the vowel sound approximated cochlear implant processing. Two blocks had them listen to the same talker each time at a spectral resolution (quality of sound) of 4 channels, the other two had them listen to a different talker each time at a spectral resolution of 8 channels. A second experiment was conducted with the same setup except with 12 and 24 channels. There was little to no interaction between spectral resolution and number of talkers, which was consistent with the hypothesis.