2026
Training Under Pressure: Designing and Evaluating a Multi-Stressor VR Environment for Public Speaking Training
Zhiyuan (Terry) Wei, Lei Zhang
ACM Symposium on User Interface Software and Technology (UIST) 2026
Effective public speaking requires performing under stress, yet prior VR training systems have primarily focused on reducing anxiety rather than building the resilience to withstand it. In this paper, we introduce a multi-stressor VR public speaking training system that combines audience non-verbal behavior, crowd size, and AI-generated contextual Q&A, each designed across manipulable levels of stress. In an exploratory within-subjects study (N = 12), we characterize how combined stressor intensity relates to speaking performance and the contribution of individual stressors. Results show that our parameterized stressor design successfully induced distinct anxiety levels, with negative non-verbal animations serving as a stronger source of stress than either crowd size or AI-generated adversarial interruptions. In addition, AI interruptions forced a cognitive context switch during speech, prompting speakers to adopt coping strategies such as adjusting eye gaze and pacing to regain composure. Finally, participants identified the high-stress negative audience condition as the most valuable for training. Together, these findings suggest that rather than minimizing stress, VR systems for public speaking should calibrate challenges to a productive zone and incorporate cognitive scaffolds that help users sustain performance under high-pressure interruptions.
Training Under Pressure: Designing and Evaluating a Multi-Stressor VR Environment for Public Speaking Training
Zhiyuan (Terry) Wei, Lei Zhang
ACM Symposium on User Interface Software and Technology (UIST) 2026
Effective public speaking requires performing under stress, yet prior VR training systems have primarily focused on reducing anxiety rather than building the resilience to withstand it. In this paper, we introduce a multi-stressor VR public speaking training system that combines audience non-verbal behavior, crowd size, and AI-generated contextual Q&A, each designed across manipulable levels of stress. In an exploratory within-subjects study (N = 12), we characterize how combined stressor intensity relates to speaking performance and the contribution of individual stressors. Results show that our parameterized stressor design successfully induced distinct anxiety levels, with negative non-verbal animations serving as a stronger source of stress than either crowd size or AI-generated adversarial interruptions. In addition, AI interruptions forced a cognitive context switch during speech, prompting speakers to adopt coping strategies such as adjusting eye gaze and pacing to regain composure. Finally, participants identified the high-stress negative audience condition as the most valuable for training. Together, these findings suggest that rather than minimizing stress, VR systems for public speaking should calibrate challenges to a productive zone and incorporate cognitive scaffolds that help users sustain performance under high-pressure interruptions.