Web-Based Multimodal Learning System to Develop Social Communication Skills

Virtual agents offer a scalable and cost-effective alternative to traditional human-led social skills training, which is often limited by the availability of professional trainers. Our web-based learning system, developed following Bellack et al.’s training model, integrates speech recognition, response selection, speech synthesis, and nonverbal behavior generation to provide automated training. To evaluate its effectiveness, we conducted a four-week study with 60 Japanese participants from the general population, focusing on four key social skills. Participants completed questionnaires assessing autistic traits, social anxiety, and changes in social communication skills post-training. Results demonstrated significant improvements, with notable results in participants’ confidence in declining requests. These findings highlight the potential of web-based virtual agents for enhancing social communication skills and suggest promising applications for social communication research and intervention programs.