2025 ASEE Annual Conference & Exposition

AI-Driven Multimodal System for Enhancing Non-Verbal Communication in Public Speaking

Presented at DSAI Technical Session 5: Educational Technology and Innovative Tools

This work-in-progress paper introduces an AI coach to train engineering students on nonverbal communication. Engineering students often find it difficult to combine verbal and nonverbal cues in their interactions to communicate their ideas effectively. Conventional pedagogical approaches in engineering place emphasis on spoken communication while ignoring the crucial function of nonverbal cues, including head movements, body language, and facial emotions. This study proposes a personalized AI-driven coaching system designed to teach nonverbal communication skills to bridge the communication gap in the engineering community. Leveraging state-of-the-art multimodal AI models, our methodology integrates a comprehensive pipeline to evaluate nonverbal cues—namely, head pose, eye contact, and facial expressions, as well as verbal cues—namely, speech transcript. For head pose analysis, we utilize the MediaPipe library to classify head orientations. Facial expression evaluation relies on models trained on the FER-2013 dataset, with dominant emotions extracted and normalized from video frames. Verbal communication is assessed using OpenAI’s Whisper for transcription and Meta’s Llama3 for context-aware evaluation. Unlike traditional classroom settings, this approach integrates advanced AI to provide personalized, real-time feedback, bridging the gap between technical and interpersonal competencies. Our AI system’s performance was evaluated against human evaluators for assessing students’ communication skills and providing personalized feedback. On the Likert scale, our AI system outperformed human evaluators by 24.25% in clarity, comprehensiveness, personalization, and specificity of the feedback.

Authors
  1. Dr. Brainerd Prince Plaksha University [biography]
  2. Siddharth Siddharth Plaksha University [biography]
  3. Hibah Ihsan Muhammad Plaksha University, Punjab
Download paper (2.37 MB)