Exploring AI-Driven Affective Avatars for Autistic Adults and Adults with Social Anxiety in Virtual Meetings

Virtual meetings often exclude individuals who rely on text-based communication, such as autistic adults and those with social anxiety.This paper introduces a prototype that converts typed text into emotive avatars using LLM technology, which convey emotional tone through modulated vocal and facial expressions.We reflect on design choices for using LLMs in accessible meetings and discuss insights from our semi-structured interviews with 18 autistic adults and adults with social anxiety.Our qualitative analysis revealed the following key insights: 1) Participants found the avatars helpful in alleviating challenges like masking and exhaustion, with some noting that the avatars enhanced their communication, increasing participation and confidence; 2) While they valued the avatars' affective capabilities, including both vocal and facial animations, they were sensitive to inaccuracies in vocal expression; and 3) Participants desired more personalized control over the avatars' affect to balance societal expectations with authentic expression.