Using sociolinguistic phonetic perception to fine tune cochlear implant simulations

Abstract

Simulations of the sound of cochlear implants (CIs) are used for people with acoustic hearing for counseling, pedagogy, and to rigorously control stimulus parameters in predictions of CI performance. The most commonly used CI simulation is the eight-channel noise vocoder, which approximates intelligibility scores in better-performing CI users. However, this method underestimates the ability of CI users to demonstrate subtle effects in phonetic perception, including the adjustment of phonetic perceptions based on talker gender (phonetic “accommodation”), which CI users can do, despite no such effect found in normal-hearing listeners when using the eight-channel noise vocoder. Toward the goal of simulating CI performance more accurately, gender-driven phonetic accommodation was measured using three styles of vocoder, including an all-channel (conventional style) noise vocoder, and a peak-picking noise vocoder that delivers the top eight channels out of 22 per time bin. Carriers were variable-bandwidth noise or a pulsatile harmonic complexes, each delivered with multiple levels of spectral resolution. CI-like phonetic accommodation effects emerged in varying degree for all vocoders except the conventional eight-channel noise vocoder, suggesting that, despite good match of overall intelligibility, it is limited in its ability to predict more subtle aspects of speech perception.