A Novel Computerized Approach to Constructing Speech Audiometry Materials: Development of a Perceptually Balanced...

Area of Science:

• Audiology
• Computational Linguistics
• Speech Science

Background:

• Standardized speech recognition tests are crucial for audiological diagnosis.
• Developing equivalent word lists across languages presents significant challenges.
• Existing methods may lack systematic approaches for balancing word characteristics.

Purpose of the Study:

• To introduce and validate a novel computerized method for creating equivalent word lists for speech recognition testing.
• To develop and demonstrate this methodology by creating consonant-vowel-consonant (CVC) word lists in Hebrew.
• To establish a systematic and objective model for speech test material development.

Main Methods:

• Phase 1: Quantified word difficulty for 275 Hebrew CVC words using 60 normal-hearing listeners.
• Phase 2: Employed a Python optimization algorithm to create 25- and 50-word lists, balancing difficulty, phonemic distribution, and familiarity.
• Phase 3: Validated list equivalency using speech recognition in quiet with 120 normal-hearing listeners.

Main Results:

• The optimization algorithm successfully generated balanced word lists.
• Validation analyses confirmed robust inter-list equivalency for speech recognition in quiet.
• The developed Hebrew lists demonstrated homogeneous psychometric functions, aligning with international standards.

Conclusions:

• A comprehensive set of validated Hebrew speech recognition test materials was developed.
• The novel methodology effectively integrates empirical difficulty measurement with computational optimization.
• This approach provides a replicable model for developing standardized speech recognition tests in diverse languages.