A fuzzy inference-based music emotion recognition system

Music is a language of emotions, and hence music emotion could be useful in music understanding, recommendation, retrieval and some other music-related applications. Many issues for music emotion recognition have been addressed by different disciplines such as physiology, psychology, cognitive science and musicology. In this paper, we focus on the challenging issue of recognizing music emotions based on subjective human emotions and acoustic music signal features and present an intelligent music emotion recognition system. In our system, music emotion recognition consists of three steps: (i) Various music features are extracted from music signal. (ii) Those music features are analyzed and mapped into some Arousal and Valence values (AV values) by a fuzzy inference engine. (iii) Music emotion is decided based on the AV values. For the accurate music emotion recognition, we invented a new user feedback method, which can minimize the emotion difference between the inference engine and user feedback. We implemented a prototype music emotion recognition system and carried out various experiments to evaluate its performance. We report some of the results.