Music boundary detection is a fundamental step of music
analysis and summarization. Existing works either use
unsupervised or supervised methods to detect boundary.
In this paper, we propose an integrated approach that
takes advantage of both methodologies. In particular, a
graph-theoretic approach is proposed to fuse the results of
an unsupervised model and a supervised one by the
knowledge of the typical length of a music section. To
further improve accuracy, a number of novel mid-level
features are develop and incorporated to the boundary
detection framework. Evaluation result on the RWC
dataset shows the effectiveness of the proposed approach.