Phoneme Recognition by Deaf Individuals Using the Multichannel Nucleus Cochlear Implant

Abstract

Experiments have been carried out to determine which cues are used in phoneme identification by deaf individuals using a cochlear implant. Five deaf individuals with a Nucleus 22-channel cochlear implant were tested with open set speech audiometry in free field without lipreading. Speech material consisted of lists of Dutch words of the Consonant-Vowel-Consonant type (CVC-words). Word scores ranged from 0 to 22%, phoneme scores from II to 54%. For each subject the responses to the initial consonant, the vowel and the final consonant were entered into separate confusion matrices. Kruskal analysis, which provided a geometric representation of these confusions, showed that in the recognition of consonants the feature of voicing is all important. Vowels were identified on the basis of the frequencies of the first and second formants. In one subject the electrode array could only partially be inserted into the cochlea, leaving roughly half the second formant area of the electrode array outside the cochlea. For this subject vowel identification was based upon the first formant and vowel duration; there was no contribution of second formant information to vowel identification. Compressing the first and second formant frequency to the limited intracochlear array did not enhance transmission of second formant information and did not improve performance. The basic findings for consonant and vowel recognition could be explained by the speech coding strategy of the Nucleus speech processor in which voicing determines stimulus periodicity and formant frequencies determine channel selection. Kruskal analysis of phoneme confusions may aid in programming and evaluating the performance of the Nucleus cochlear implant.

Key Words:

• cochlear implant
• speech audiometry
• Kruskal analysis
• speech features