Effects of simulated cochlear-implant processing on voice quality distinction: Evidence from analysis of disordered voices

Abstract

<jats:p>The ability to discern variations in talkers voice quality is important for effective talker identification and robust speech processing; yet, little is known about how faithfully acoustic information relevant to variations in talkers voice quality is transmitted through cochlear implant (CI) speech processing. This study analyzed unprocessed and CI-simulated versions of sustained vowel sounds /a/ from two groups of individuals with normal and disordered voice qualities to investigate the effects of CI speech processing on acoustic information relevant to the talkers voice quality distinction. The CI-simulated stimuli were created by processing the vowel sounds using 4-, 8-, 12-, 16-, 22-, and 32-channel noise-vocoders. The voice quality for each stimulus was characterized by calculating mel-frequency cepstral coefficients (MFCCs). Then, the effects of CI speech processing on the acoustic distinctiveness between normal and disordered voices was measured by calculating the Mahalanobis distance and classification accuracy of support vector machines (SVMs) on their MFCC features. The results showed that CI noise vocoding is substantially detrimental to acoustic information involved in voice quality distinction, suggesting that CI listeners likely experience difficulties in perceiving voice quality variations. The results underscore challenges CI users may face for effective recognition of talkers and processing their speech.</jats:p>