Music mood tracking based on HCS

Music mood present the inherent emotional state of music on certain duration of music segment. The mood may vary in an entire piece of music, thus tracking the mood changing is an effect way to automatically analyze the music mood for better music understanding. HCS (Hierarchical Classification Scheme), which automatically generates tree structure hierarchical classifier driven by training data to adaptive different recognition applications, is proposed in our paper to build recognition model of music mood. The duration of training music segments is set as 8 seconds, which is assumed to contain stable mood state within a single segment to ensure the accuracy of the HCS model. The mood in an entire piece of music is then divided into 50% overlapping 8 seconds frames, which are the same duration with the training segments in HCS model, to accomplish the mood tracking. The automatic tracking result show good accordance with general music reviews and manually labeled ground truth, and proves the effectiveness of the HCS scheme and the rational selection of 8 seconds segment duration.