Vowel identification and pitch perceptions by cochlear implant users

Lai, T.T., Svirsky, M.A., Meyer, T.A., Kaiser, A.R., Basalo, S, Silveira, A., Suárez, H., Simmons, P.M. & Miyamoto, R.T. (1999). Eastern Society for Pediatric Research. Journal of Investigative Medicine, 47 (2) Feb. 1999. Atlantic City, NJ.

 

Abstract

Despite advances in implant technology, cochlear implant (CI) users demonstrate a wide range of the ability to perceive speech in the absence of visual cues. To explain speech perception with a CI, Svirsky & Meyer (1997) developed a psychophysically-based mathematical model [Multidimensional Phoneme Identification (MPI)]. The model provides a full description of how listeners encode, represent and combine the sensory information elicited by vowel sounds. The model generates predicted confusion matrices, given a listener’s performance on various psychophysical tasks. The model predicts, for any given listener, how well s/he perceives vowels, as well as which pairs of vowels are and are not confused by the listener. A three-dimensional model has successfully predicted vowel identification for a group of experienced CI users. In this study, we were interested in using the MPI model to predict long-term vowel perception for an individual CI user based on performance on a psychophysical (pitch ranking) task at the time of initial stimulation. Vowel perception improves with continued CI use. Is this improvement due to increases in the listener’s ability to discriminate between sounds or to label vowel sounds according to a learned pattern? We measured vowel identification and pitch ranking at the time of initial stimulation and several months later. Vowel identification was obtained using ten to fifteen repetitions of each Spanish vowel (I,E,A,O,U) presented in a /hVd/-like context. In the pitch ranking task, the subject heard stimuli sent to adjacent pairs of intra-cochlear electrodes and had to indicate which stimulus was higher pitched. A software version of the MPI model was implemented to obtain predicted vowel confusion matrices based on the patient’s pitch ranking capabilities. The subject’s ability to accurately rank the pitch of different intracochlear electrodes did not vary much between initial stimulation and 6 months of use. Vowel identification, however, increased from approximately chance performance (24%) at initial stimulation to 86% correct identification by 4 months of implant use. Based on the subject’s pitch ranking ability at initial stimulation, the model generated a reasonably accurate prediction of her vowel identification performance 4 months post-implant. The results suggest that the MPI model is an appropriate tool for predicting vowel perception in individual CI users, and that improvements in vowel identification are related to changes in vowel labeling rather than to changes in pitch ranking abilities.