Modeling Affective Responses to Music Using Audio Signal Analysis and Physiology

摘要

A key issue in designing personalized music affective applications is to find effective ways to direct emotion by music selection with appropriate combination of acoustic features. The aim of this study is to understand the dynamic relationships between acoustic features, physiology and affective states. To model these relationships we used a multivariate approach including continuous measures of emotions from behavioral, subjective and physiological responses. Classical music excerpts taken from opera overtures were used as stimuli to induce emotional variations across time between neutral and intense emotional states. Continuous ratings of arousal and valence along with cardiovascular, respiratory, skin conductance and facial expressive activity were recorded simultaneously. Results show that parts of the music with higher loudness and pulse clarity induced higher ratings of arousal, sympathetic activation and increased cardiorespiratory synchronization. In contrast, pleasant and calming parts with major mode and prominent key strength induced higher ratings of valence, parasympathetic activation and increased facial activity.