Test-Retest Reliability Bounds for Consonant-Nucleus-Consonant Word and Phoneme Scores for Cochlear-Implant Users

Area of Science:

• Audiology
• Speech-Language Pathology
• Biostatistics

Background:

• Cochlear implants (CIs) aim to restore hearing but recognition abilities vary.
• Establishing reliable measures for speech recognition is crucial for assessing CI efficacy.
• Consonant-vowel-consonant (CNC) tests are standard for evaluating speech perception.

Purpose of the Study:

• To determine test-retest 95% confidence interval bounds for CNC word and phoneme recognition scores in CI listeners.
• To model the effective set size for word and phoneme recognition based on listener data.

Main Methods:

• Analysis of data from two existing cochlear implant studies.
• Comparison of across-list standard deviations (SDs) with expected binomial distribution SDs.
• Adjustment of the binomial distribution model by tuning the effective set size to minimize deviations from measured SDs.

Main Results:

• The effective set size for CNC words matched the number of words per list (50).
• The effective set size for phonemes was approximately 100, lower than the 150 phonemes presented per list.
• This indicates mutual information across phonemes during recognition.

Conclusions:

• Results support the hypothesis of independent recognition for individual CNC words.
• Evidence suggests interconnectedness (mutual information) across phonemes, leading to a smaller effective set size.
• A table of confidence intervals and critical-difference values for CNC word and phoneme recognition is provided for clinical use.