Child-therapist acoustic synchrony and response trajectories in autism intervention: an AI-based automated analysis using dynamic systems theory and affective computing

• AI-based pipeline models child-therapist acoustic synchrony in autism intervention • Explores short- and long-term nonlinear, complex dynamics • Synchrony metrics specific to intervention phases are linked to response • Prosodic features reflect affective content relevant to therapy outcomes • Novel, non-invasive method to monitor therapy dynamics at scale Child-clinician interpersonal dynamics are central to psychotherapy and are increasingly acknowledged as key elements in autism intervention. However, quantitatively studying fine-grained aspects such as the child-clinician synchrony patterns poses challenges, limiting translational research. Moreover, synchrony is rarely investigated with a long-term perspective. This study employed an AI-based, fully automated computational pipeline to analyze child-clinician interpersonal acoustic synchrony through the lens of complex dynamic systems and affective computing. We followed 25 autistic preschoolers over one year of Naturalistic Developmental Behavioral Intervention (NDBI). Three 60-minute intervention sessions, at the beginning, after three months, and after one year, were analyzed second-by-second, totaling 75 videos. After AI-based automatic speech segmentation, acoustic synchrony was assessed using Cross-Recurrence Quantification Analysis to derive interaction metrics over the entire therapy sessions employing affective prosodic features. Robust Bayesian correlation analysis was used to explore the relationship between affective acoustic synchrony and developmental learning rates at different time points. No significant associations were found at baseline, while correlations emerged after three months and became more pronounced at one year. Early in therapy, interactions with a stronger internal structure, particularly in loudness, spectral dynamics, and voice quality, were linked to higher developmental gains. After one year, the relationship between synchrony and response shifted toward metrics reflecting transition dynamics and stability. Associations with fine-grained spectral features particularly characterized this phase. Specific and different synchrony aspects were associated with therapy response trajectories both in the initial and latter phases of therapy. Acoustic features involved in intervention response are known to participate in the emotional content of speech, highlighting the contribution of affective aspects to therapy. These findings provide valuable insights into the role of interpersonal synchrony in autism intervention and underscore the potential of computational methods in monitoring treatment progress.