A multimodal read of a ~40-minute dyadic conflict: vocal arousal for each partner over time, the phases it moved through, who interrupted when, and how tightly the two arousal levels tracked each other.
Each line is scaled to its own speaker's baseline (within-speaker z-score), so the comparison is change over time, not who was louder. Coupling is a sliding-window correlation between the two arousal series: above the line, the partners rise and fall together. Hover for values.
Near-symmetric dynamics. Talk time, turn-taking, repair attempts, and arousal are closely matched between partners, and once the conversation engaged, their arousal moved together. The pattern reads as a mutual cycle, not a one-directional one.
Clinical priority — read before the interaction metrics
This session contains markers of individual distress, including expressed self-directed risk during the distress phase. Individual safety and wellbeing take precedence over any interaction-pattern work shown below. This dashboard is a descriptive aid, not a risk assessment or diagnosis.
Keyword-family proxies, counted per speaker — not validated clinical codes, and counts rather than severity. Both partners show the same patterns; the difference is degree.
Each node is a clinically relevant indicator. An edge appears when two move together within a ~5-minute window around the current time (blue = together, amber = opposite; |r| ≥ 0.35), and node size reflects that indicator's level in the window. Press Play to watch the structure reorganize across the conversation. Exploratory: descriptive associations from a single session — illustrative, not estimated model parameters.