An Automatic Segmentation Method of Popular Music Based on SVM and Self-similarity

Popular music segmentation plays an important role in popular music content retrieval, melody extraction and semantic understanding etc. Segmentation boundary detection is one of the key technologies in lots of conventional algorithms. In this paper, we propose an automatic segmentation approach that combines SVM classification and audio self-similarity segmentation, which firstly separate the sung clips and accompaniment clips from pop music by using SVM preliminary classification, and then heuristic rules are used to filter and merge the classification result to determine potential segment boundaries further. Finally, self-similarity detecting algorithm is introduced to refine our segmentation results in the vicinity of potential points. Experiment shows that our approach can achieve more sophisticated and accurate results by using the local change context substantially to determine the segmentation boundaries.