Classroom-based speech recognition better predicts listening comprehension than clinic-based measures in primary school children.

Classroom noise can substantially interfere with children's access to spoken instruction, yet speech perception is typically assessed using clinic-based measures collected under controlled listening conditions. This study examined (1) how classroom signal-to-noise ratio (SNR) and developmental level influence speech recognition (SR) and listening comprehension (LC), and (2) whether clinic-based SR predicts classroom LC as effectively as classroom-based SR. One hundred eleven normal-hearing students in Grades 2-5 completed SR (WIPI) and LC (TROG-2) tasks in occupied classrooms under multiple noise levels, as well as SR testing in a clinic setting. Generalized linear mixed-effects models were used to estimate acoustic and developmental effects, and receiver operating characteristic analyses compared predictive accuracy. Higher SNR was associated with significant improvements in both SR and LC, and older students consistently outperformed younger students across acoustic conditions. Although SR performance was higher in the clinic than in the classroom, classroom-based SR provided substantially better prediction of classroom LC than clinic-based SR. These findings demonstrate that assessment context shapes observed listening performance and underscore the importance of ecologically valid measures for predicting functional comprehension outcomes in real educational environments.