Institute of Information Science, Academia Sinica

This talk provides a comprehensive introduction of the research on modeling human's emotion perception of music, a research topic that emerges in the face of the explosive growth of digital music. Automatic recognition of the perceived emotion of music allows users to retrieve and organize their music collections in a fashion that is more content-centric than conventional metadata-based methods. Building such a music emotion recognition system, however, is challenging because of the subjective nature of emotion perception. One needs to deal with issues such as the reliability of ground truth data and the difficulty of evaluating the prediction result, which do not exist in other pattern recognition problems such as face recognition and speech recognition. This talk provides the details of the methods that have been developed to address the issues related to the ambiguity and granularity of emotion description, the heavy cognitive load of emotion annotation, the subjectivity of emotion perception, and the semantic gap between low-level audio signal and high-level emotion perception.