EmotiCon: Enhancing Public Speaking Training with Context-Aware Feedback on Emotional Delivery

Abstract

Effective public speaking performances are significantly influenced by the alignment of a speaker's emotional delivery with the semantic content of their speech. While current digital coaches for public speaking have made strides in analyzing paralinguistic features, they often overlook the underlying context of what is being spoken. This paper introduces EmotiCon, a novel system designed to bridge this gap by providing context-aware feedback on emotional delivery. EmotiCon employs a multimodal pipeline that analyzes the user's speech audio, the transcribed text, and the source material to assess the congruence between vocal tonality and semantic meaning. The system segments the speech, performs emotion recognition on the audio and transcribed text, and retrieves the most relevant contextual passages from user-provided documents. By comparing the emotional sentiment across these three modalities, EmotiCon generates specific, actionable recommendations using a Large Language Model to help users enhance their delivery. We conducted a study with 8 higher education students to evaluate the system. The findings from this study, based on user interaction with a visual feedback dashboard and semi-structured interviews, indicate the potential of context-aware feedback to improve pedagogical effectiveness.