Content-based music structure analysis

Abstract

This thesis proposes a framework for popular music structure detection, which incorporates music knowledge with audio signal processing techniques. The important components of the music structure (timing, harmony, music regions and semantic meanings) are modeled hierarchically in the layers of the music structure pyramid. In our framework, we first propose a rhythm based music segmentation technique to segment the music. Then octave varying temporal characteristics are taken into account to design the features to characterize the music signals. Finally we optimize the parameter of the existing statistical learning techniques for music signal modeling. Our experimental results comparison shows that the incorporation of music knowledge to audio signal processing can more effectively detect the music structure than using conventional speech processing techniques. Our ideas on how this structural information can potentially be used for applications are then discussed in the thesis.