| Summary: | Purpose: The purpose of this study was to introduce and validate a novel computerized approach for constructing equivalent word lists for speech recognition testing and to demonstrate this methodology through the development of consonant–vowel–consonant (CVC) word lists in Hebrew. Method: The study was conducted in three phases. Phase 1 empirically quantified word difficulty (50% recognition threshold in noise) for 275 Hebrew CVC words among 60 normal-hearing listeners. Phase 2 utilized a custom Python optimization algorithm to allocate 200 words into eight 25-word lists and four 50-word sets. The algorithm simultaneously balanced multiple variables, including perceptual difficulty (50% point), phonemic distribution, and word familiarity. Phase 3 empirically validated the lists’ equivalency for speech recognition in quiet among 120 normal-hearing listeners by establishing performance–intensity functions. Results: The optimization algorithm successfully generated balanced lists. Validation analyses of speech recognition scores in quiet (analyses of variances) demonstrated robust interlist equivalency; no statistically significant differences were found among the 25-word lists or 50-word sets at any of the presentation levels tested. The lists exhibited homogeneous psychometric functions (e.g., mean slope of 4.64%/dB for 50-word sets) consistent with international standards. Conclusions: A comprehensive set of Hebrew speech recognition test materials was successfully developed and validated. The novel methodology, integrating empirical difficulty measurement with computational optimization, proved effective. This approach provides a systematic, objective, and replicable model for developing standardized speech recognition tests across diverse languages.
|
|---|