On the relationships between music-induced emotion and physiological signals

Abstract

Emotion-aware music information retrieval (MIR) has been difficult due to the subjectivity and temporality of emotion responses to music. Physiological signals are regarded as related to emotion and thus could potentially be exploited in emotion-aware music discovery. This study explored the possibility of using physiological signals to detect users’ emotion responses to music, with consideration of individual characteristics (personality, music preferences, etc.). A user experiment was conducted with 23 participants who searched for music in a novel MIR system. Users’ listening behaviors and self-reported emotion responses to a total of 628 music pieces were collected. During music listening, a series of peripheral physiological signals (e.g., heart rate, skin conductance) were recorded from participants unobtrusively using a research-grade wearable wristband. A set of features in the time- and frequency- domains were extracted from the physiological signals and analyzed using statistical and machine learning methods. Results reveal 1) significant differences in some physiological features between positive and negative arousal and mood categories, and 2) effective classification of emotion responses based on physiological signals for some individuals. The findings can contribute to further improvement of emotion-aware intelligent MIR systems exploiting physiological signals as an objective and personalized input.