Automated Multilingual Naturalistic Speech Comprehension Assessment Using LLMs

This study develops a novel automated method leveraging large language models (LLMs) to generate and score naturalistic spoken story recall across multiple languages. By testing native speakers of English and 10 other languages in quiet and babble noise conditions, the approach demonstrated accurate, language-independent scoring sensitive to known cognitive effects. This fully automated system overcomes traditional speech test limitations, offering a scalable, clinically relevant tool for assessing real-world speech comprehension.