AI-Driven Analytics of Team-Teaching Talk: Acoustic Patterns across Experience, Cohorts and the Learning Design

Synthesis: This paper presents an AI-based speech processing approach to analyze classroom talk in team-teaching settings, grounded in spatial pedagogy theory. Analyzing 36 recorded undergraduate and postgraduate sessions involving 12 teachers, the study extracts acoustic features (voice quality, intonation, loudness) and codes spatial pedagogy behaviors. Results reveal systematic differences most notably in loudness dynamics: high-experience teachers, undergraduate classes, and Collaborative Learning tasks exhibited greater loudness variation, suggesting more frequent modulation of volume to foreground key information and support engagement. This is the first large-scale automated analysis of acoustic patterns in team-teaching, demonstrating that AI can scalably capture meaningful teaching-talk features across experience, cohort, and task design. The work contributes to Learning Analytics by extending classroom sensing beyond student-focused clickstream data to teacher vocal behavior, and to Teaching research by showing how AI can provide formative feedback on teaching practice.

What this means for practice

• Instructors. Treat loudness variation as a teachable vocal strategy: high-experience teachers showed greater loudness variation than less experienced colleagues, and deliberate modulation of volume foregrounds key information and signals transitions across lesson phases.
• Faculty developers. Use automated speech analytics to generate Feedback on teaching talk — the pipeline analyzed 34 of 36 recorded sessions in authentic team-taught classrooms, a scale manual transcription cannot reach when multiple teachers speak across extended sessions.
• Instructors. Plan for more vocal modulation during collaborative learning tasks and in undergraduate sessions, where loudness dynamics were largest, and reflect on which moments those shifts marked.
• Instructors. Treat the acoustic output as formative input on teaching practice rather than an evaluation, since the features indicate how talk was enacted, not its pedagogical quality.

Limitations

• The data come from 36 recorded sessions in one undergraduate and postgraduate database course at a single institution over four consecutive weeks, a single institutional and language context the authors identify as limiting generalizability.
• Only 34 sessions entered the analysis: a microphone placement issue made one instructor's audio unusable in two sessions.
• Acoustic measures were derived from classroom audio alone, with no gaze, gesture, or spatial positioning data, so explanations of why loudness variation occurs remain inferential.
• The unit of analysis was the individual teacher, which the authors note limits inferences about team-level coordination even though it establishes that experience-related acoustic differences are detectable.

Citation

Liu, Y., Martinez-Maldonado, R., Alfredo, R., Mejia-Domenzain, P., Rahayu, D., & Nawaz, S. (2026). AI-Driven Analytics of Team-Teaching Talk: Acoustic Patterns across Experience, Cohorts and the Learning Design.