As the prevalence of autonomous interactive agents is growing incredibly fast, it becomes increasingly clear that these virtual agents must not only comprehend and respond to our verbal content but also engage with our emotions, which is crucial for enabling more profound interactions. While recent advancements in AI have significantly improved the automatic recognition and understanding of human speech, challenges persist in accurately identifying and addressing the nuances of human emotions. 

We assume that an empathic virtual agent should excel in at least three key tasks: i) recognising human's spontaneous emotional expressions alongside understanding the verbal content, ii) generating appropriate responses in terms of timing and style, and iii) providing insightful feedback while comprehending user responses.  In order to accelerate the development of empathic agents, we introduce the first Empathic Virtual Agent Challenge: EVAC. In its inaugural edition, the focus is set on the robust recognition of spontaneous human expressions during interactions with a virtual agent, using the recently introduced THERADIA WoZ dataset. Participants will have to predict the intensity of dimensional or categorical attributes of affect, from audiovisual sequences of human interactions in French with a virtual agent. We encourage the participation from both academics and the industry.