User-adaptive music emotion recognition

Music can arouse profound and deep emotional reactions and the automatic emotion recognition of music is useful for music information retrieval, human-computer interaction and affective computing Picard R.W. (1997). However, the nature of music is very complex and users' emotion responses vary from individual to individual. In this paper, we present an adaptive scheme to recognize the emotional meaning of music, which is able to follow users' preference. The recognition process is consisted of four steps: first, a two-dimensional model, 'emotion plane' is used to model the emotion classes; second, novel musical perceptual features are extracted from MIDI files; then, different support vector machines (SVM) were trained according to different users' preferences and finally these trained support vector machines are used to classify the emotion of music. Satisfying experimental results are obtained on western tonal music, with different users. That indicates the effectiveness of our approach.