Facial expression-based assessment of emotional engagement under multimodal stimuli

This study investigates the feasibility of an emotion-derived engagement proxy computed from facial expression recognition (FER) outputs in response to auditory and audiovisual stimuli, in an immersive and acoustically isolated setting relevant to experience evaluation. Thirty-seven participants were exposed to three stimulus blocks: Sound, Sound with Video (same audio content and duration), and a validated strong-elicitation benchmark (FilmStim excerpts). The Sound vs. Sound with Video contrast was used as an expected sensitivity check, while FilmStim served to contextualise response magnitude. Within the fixed-sequence protocol, the proposed FER-derived engagement proxy discriminated the three blocks in the expected direction. We discuss limitations related to fixed order, demographic composition, and FER uncertainty, and outline how the approach can complement self-report and sensor-based measures in early-stage evaluation workflows.