Rhythm-Based Segmentation of Popular Chinese Music

We present a new method to segment popular music based on rhythm. By computing a shortest path based on the self-similarity matrix calculated from a model of rhythm, segmenting boundaries are found along the diagonal of the matrix. The cost of a new segment is optimized by matching manual and automatic segment boundaries. We compile a small song database of 21 randomly selected popular Chinese songs which come from Chinese Mainland, Taiwan and Hong Kong. The segmenting results on the small corpus show that 78% manual segmentation points are detected and 74% automatic segmentation points are correct. Automatic segmentation achieved 100% correct detection for 5 songs. The results are very encouraging.

• Type: Conference
• Year: 2005
• Where: ISMIR
• Authors: Kristoffer Jensen, Jieping Xu, Martin Zachariasen
• Keywords: Automatic Segment Boundaries, Automatic Segmentation, Information Retrieval, ISMIR 2005, Segmentation Points