Developing finer-grained methods for analyzing word recognition performance by individuals with cochlear implants

Abstract

Cochlear implant (CI) listeners experience overall improvement in speech recognition abilities, but a considerable amount of variability still remains among these individuals. Data from simulated CI listening conditions also show wide variability in monosyllabic word recognition ability. In an adverse simulated condition (2 spectral channels using sinusoidal carriers), the word score can be nearly 0 for some adults with normal hearing. These floor effects pose a challenge for researchers, as they lead to data compression. When analyzing the performance by phonemes, the increase in scores may remove the floor effect. The purpose of this study is to establish the chance performance rate for NU-6 consonant-vowel-consonant (CVC) word lists analyzed by phonemes. To determine this, we randomly paired words from the NU-6 list with CVC words from an on-line corpus of English, the English Lexicon Project, and scored the matches by phonemes using dictionary pronunciations. Thousands of runs of these random pairings were used to generate expected baseline accuracy ranges. These results were compared with individual, behavioral data from people listening to NU-6 words processed using 2-channel sinusoidal vocoders to determine if participants were performing better than chance when scoring by phonemes.