What Sounds Dangerous? Establishing Correlations Of Musical Features and Perceived Safety in HRI

Ahstract- This study explores the potential of music driven sonification as an effective method for improving safety in humanrobot collaboration. Building on the rich expressive content of music, this study assesses the communicative potential of both low level musical features, such as pitch, tempo, and timbre; and high level music features of rhythmic stability and tension-release. Two music datasets have been created, labeled, and evaluated based on five criteria: safety/danger, approachability, risk of failure, and urgency. The first dataset consists of prerecorded song clips while the second one contains original compositions designed to isolate high-level musical features. 400 participants annotated our datasets base on the five criteria. Our findings reveal significant correlations between musical features such as timbre, harmonic tension, and note onset; and the perception of safety, urgency, and risk. Based on these results, we developed a framework and an audio plugin for music-driven sonification of robotic gestures to support safe human-robot interaction.