ENERGI: a multimodal data corpus of interaction of participants in virtual communication

A statistical analysis of the multimodal ENERGI (ENgagement and Emotion Russian Gathering Interlocutors) data corpus containing audio-video recordings of communication in Russian by a group of people obtained using the Zoom teleconference system has been performed. The corpus data is annotated into three classes: participant engagement (high, medium, low), emotional arousal (high, medium, low), and emotional valence (positive, neutral, negative), as well as ten classes of communicative gestures. The corpus contains 6.4 hours of video recordings of group communications, with a total of 18 unique speakers; the data is annotated using 10-second time intervals. ENERGI’s advantages over other corpora include its multimodality, Russian language support, speaker diversity, natural recording conditions, and extensive annotation across several behavioral parameters of communication participants. The corpus can be used to develop a multimodal automated system for analyzing the behavioral aspects of participants in virtual group communications.