Tri-word presentations with phonemic scoring for practical high-reliability speech recognition assessment.

Area of Science:

• Audiology
• Speech-Language Pathology
• Psychometrics

Background:

• Speech recognition test reliability is crucial for accurate assessment.
• Traditional tests may not always meet practical clinical constraints.
• The Computer Assisted Speech Recognition Assessment (CASRA) offers a flexible approach.

Purpose of the Study:

• To evaluate the empirical reliability of shortened CASRA versions (10, 20, and 25 sets).
• To determine how well shortened CASRA versions predict full-test scores.
• To assess the test-retest reliability of shortened CASRA versions for diverse hearing abilities.

Main Methods:

• Administered 10-, 20-, and 25-set CASRA versions to 144 participants.
• Collected test and retest scores for each version.
• Analyzed group data for correlations and variance.
• Examined individual test-retest reliability using binomial confidence intervals.

Main Results:

• Group test-retest scores were highly correlated and not significantly different across all tested CASRA versions.
• 20- and 25-set versions predicted approximately 97% of the full-test score variance.
• The 10-set version predicted about 88% of the full-test score variance.
• Individual reliability showed acceptable performance, with fewer than 7.6% of cases outside 95% confidence limits.

Conclusions:

• Shortened CASRA versions provide a practical method for improving speech recognition test reliability compared to traditional word tests.
• The 20-set version offers a strong compromise between test length and reliability for clinical use.
• The 10-set CASRA version is recommended for routine use due to its practicality, reliability, and minimal learning effects.