How gender shapes public speaking in virtual reality: a multimodal analysis

Virtual Reality (VR) has proven effective for public speaking training, particularly due to its ability to simulate immersive speaking environments and automatically extract behavioral cues reflecting performance. However, limited attention has been paid to the impact of speaker gender and the gendered appearance of the virtual audience on the public speaking experience in VR. To address this gap, we conducted an experiment with 55 participants who performed oral presentations within a VR simulation that allowed for the systematic manipulation of the virtual audience's gendered appearance and social attitude. We captured a comprehensive dataset of multimodal nonverbal behavioral data, alongside subjective evaluations from both the participants and an expert using standardized and non-standardized questionnaires. By combining statistical analysis and machine learning approaches, our findings revealed no significant effect of speaker gender on her/his emotional experience or her/his public speaking performance. In contrast, male and female speakers exhibited distinct nonverbal behavioral patterns, identified through both expert evaluation and automatic multimodal behavioral analysis. Moreover, Virtual audience social attitude influenced participants' emotional experience and self-reported performance, whereas no significant effect of audience gendered appearance was detected. These findings highlight the importance of considering behavioral variability when studying and assessing public speaking in VR.