Examining hearing-aid benefit during story listening in older adults using large language models

Hearing-aid benefit is typically assessed with isolated sentences, whereas everyday listening involves continuous speech. Yet, scoring comprehension of such speech can be labor-intensive. Here, we examined whether large language models (LLMs) can assess speech intelligibility and story comprehension in older hearing-aid users, and identify who benefits from hearing aids. Thirty-nine older adults (median age: 75 years) listened to sentences and short stories in multi-talker babble, under aided and unaided conditions. LLM-based intelligibility scores were nearly identical to established, rule-based scores. LLM-based intelligibility, gist, and detail recall scores were sensitive to masking and hearing-aid use. On average, hearing aids improved sentence intelligibility and story recall. Benefits increased with poorer hearing and, independently, with greater audibility gains from hearing aids. However, about half of the participants showed little benefit, primarily those with milder (though still clinically significant) hearing loss or smaller improvements in audibility. The results demonstrate that benefits of hearing aids for sentence intelligibility extended to comprehension of short stories and that LLMs enable automated assessment of both.