Abstract: The prevalence of online reading platforms has surged along with improvements to automatic speech recognition (ASR) systems. These platforms facilitate personalized reading instruction, but like all learning systems, work best when students find them engaging. The Educational Data Mining (EDM) community has made great efforts to detect student engagement, but these efforts have largely been confined to school and lab environments without intentionally including neurodiverse populations.
In the present study, we evaluate engagement levels in children with attention deficit hyperactivity disorder (ADHD, n=55), autism (n=38), both autism and ADHD (n=38) and with no known clinical diagnoses (n=22) as they interact with a digital literacy learning system from their homes. Engagement was measured using parent report, and acoustic features were extracted from audio recordings of the students while they interacted with the Amira literacy learning system.
Our results indicate that acoustic measures related to temporal features, pitch, and vocal quality are related to engagement levels across all clinical groups. We also identify key temporal features that vary across different clinical groups. These findings demonstrate the validity of using non-invasive acoustic features to measure engagement levels during reading comprehension tasks at home. However, interaction effects also highlight the importance of taking into account children’s clinical diagnosis when measuring engagement.
Suggested citation: Perkoff, E. M., Lyons, M., Mulqueen, A., Baker, R. S., Gee, A., Ocumpaugh, J., Porsch, A., Ryant, N., Parish-Morris, J., & Ungar, L. (2026). Engagement-driven differences in acoustic features during online reading comprehension conversations. In Proceedings of the 19th International Conference on Educational Data Mining. International Educational Data Mining Society.