An exploratory study of AI-avatar speaking practice in low-immersion VR: Fluency, anxiety, and technology acceptance

Abstract

This exploratory mixed-methods study investigated whether speaking practice with AI-controlled avatars in low-immersion virtual reality (LiVR) on personal computers is associated with improved speaking fluency and reduced foreign language anxiety (FLA) among Japanese university EFL learners, and how learners perceive the platform. Eight students practiced speaking with AI-controlled avatars in Immerse at home for one month. Data included pre/post FLA surveys, audio recordings, reflective journals, and a post Technology Acceptance Model (TAM) survey. Results showed a small but statistically significant increase in fluency (words per minute) and a modest, statistically significant decrease in FLA. Qualitative data and TAM responses suggested LiVR practice was engaging, convenient, and emotionally supportive, while highlighting limits in conversational flexibility and occasional recognition problems. The fluency-FLA association was moderate but not statistically significant. Given the very small sample (N = 8), short duration, lack of a control group, and a single fluency metric, findings should be interpreted cautiously. Future controlled studies with larger samples and multidimensional fluency measures are needed to test causal effects and refine links among fluency, anxiety, and acceptance. As VR and generative AI evolve, AI-controlled avatars in LiVR may expand accessible interaction opportunities in EFL settings with limited authentic practice.