Samples-based automatic key segment extraction for popular songs

In this paper, we define the key segment of a popular song as the most impressive part in this song. It can be regarded as a perfect index or summary for this song, therefore it is much helpful for popular song's audition and management. We propose an effective algorithm for automatic key segment extraction based on the analysis of positive and negative samples of key segment extracted manually from a training database. An experiment is carried on a test database including 130 popular songs. Compared with manually extraction of key segments, the proposed algorithm has a rather satisfying performance.