Item

Abstract

Music-evoked chills (MECs) are physiological responses to pleasurable events in music. Existing research on properties of music that elicit MECs has focused on low-level acoustic features in small samples of music. We created a large dataset of over 1,000 pieces of music timestamped with MECs and used computational methods to predict MEC onsets from both low-level acoustic features and high-level musical expectations. A machine learning classifier was trained to distinguish MEC onsets from non-MEC passages in the same pieces. The results show that MEC onsets are predicted better than chance and corroborate evidence for acoustic elicitors of chills with a much larger dataset. They also produce new empirical evidence that MECs are elicited by expectation, which is a more effective predictor of MEC onsets than acoustic elicitors, and may generalise to pleasurable experience in other domains such as language comprehension or visual perception.