Acoustic and Facial Markers of Perceived Conversational Success in Spontaneous Speech

A research team from the University of Maryland, led by Thanushi Withanage, Elizabeth Redcay, and Carol Espy-Wilson, has published a study analyzing what makes Zoom conversations successful. By applying multimodal AI to a large dataset of spontaneous, non-task-oriented video calls, they moved beyond traditional lab settings to capture real-world interaction dynamics. The AI extracted quantifiable features across two key modalities: acoustic signals (like speech pitch and intensity) and visual signals (like facial movements and turn-taking patterns).

The core finding is that 'entrainment'—the phenomenon where conversation partners subconsciously synchronize their speaking patterns—is a reliable marker of perceived success, even in casual, virtual settings. Perceived success was measured through post-conversation ratings analyzed via factor analysis. This means specific, measurable behaviors (e.g., mirrored pause lengths, aligned vocal pitch ranges) directly correlate with how positively participants rated their interaction.

This research, accepted for presentation at the prestigious ICASSP 2026 conference, provides a data-driven framework for understanding communication quality. It identifies key interactional markers that AI systems can detect in real-time. The study highlights significant opportunities for developing targeted interventions, such as AI-powered communication coaches or meeting analytics tools that provide feedback to foster more effective and engaging virtual dialogue.