Assessing the Risk of Aggression Using Vocal Biomarkers

Workplace violence and aggression remain persistent challenges in healthcare, where earlier awareness of escalating risk may give care teams more opportunity to de-escalate situations and protect staff, patients, and visitors. The Canary Safety Model explores whether speech can provide a continuous, ambient signal of elevated aggression risk. The model combines four vocal biomarker components-anxiety, vocal power (loudness), speaking speed, and vocal dynamics-each supported by existing research connecting measurable characteristics of speech with anger, arousal, dominance, and aggressive behavior. Designed to work with conversational speech rather than specified prompts or read speech, the model can continuously process speech and generate a dynamic room-level risk signal.

Canary applied the model to speech samples from 981 individuals and found that 71% fell into the low-risk category, 27% into the medium-risk category, and 2% into the high-risk category. An informal listening review of a subset of high-risk samples found that many exhibited distinctly aggressive tone or content, providing qualitative support for the highest risk range. These findings support the potential of vocal biomarkers as an additional source of situational awareness in healthcare settings, complementing existing observation-based assessments and emergency response systems. The Canary Safety Model is designed to provide a continuously updating signal that can inform established workflows for de-escalation and healthcare worker protection.