Biomedical Engineering graduate research seminar
Polina Stepanenko, MASc candidate and Ben Masters, PhD candidate, both supervised by Professor Ewen MacDonald, will present their research.
Abstract
Hearing aids are typically evaluated in isolated listening tasks, which do not account for the added cognitive load of speech planning required for fluid communication. Assessing the performance of hearing aids during conversation requires reliable measures of communication difficulty, though. Previous work has demonstrated that behavior, such as the timing of turn-taking, can be used to infer effort during conversations. Here we present two new complementary approaches for measuring effort during communication:
Physiological approach presented by Ben Masters
The size of the pupil is known to increase with cognitive workload. We demonstrate that this remains true in conversation and show that pupil size changes consistently around turn-taking in a manner indicative of effort. Further, background noise was shown to increase the size of pupil dilations at turn-transitions, and hearing loss resulted in a sustain of these dilations. Similar patterns of responses are observed across different experiments, populations, and cohorts, suggesting that this could act as a reliable measure in specific conversational contexts.
Linguistic approach presented by Polina Stepanenko
While prior work has investigated how the timing of speech changes with difficulty, we also expect the content of speech to change. This work examines how content-level measures, lexical diversity and semantic similarity, vary across listening and task conditions. We apply Natural Language Processing techniques to a dataset of native English speakers conversing in quiet and in noise, under both free and task-based conditions. We found that background noise reduced semantic dispersion, suggesting that listeners focused more narrowly and engaged in fewer off-topic shifts. We analyze these measures at the turn level, alongside conversational timing, to capture patterns of lexical diversity and semantic similarity across conversation. This approach allows us to examine whether adaptations to adverse conditions emerge at the local (turn) or global (conversation) scale, and whether speakers adjust their linguistic strategies dynamically in response to immediate conversational demands.
BME Grad Students: you are required to attend an average of four seminars per term as part of your degree requirements.