A FAST ALGORITHM FOR AUTOMATIC KEY SEGMENT EXTRACTION FOR POPULAR SONGS

摘要

The key segment of a popular song is defined as the most impressive part of this song. Usually, it is the climactic part of a song. It can be regarded as a perfect summarization of a song, and can be used for the purpose of music audition, retrieval, and management. In this paper, a fast algorithm for automatic key segment extraction for popular songs is proposed. First, the audio segments with relatively high energy and proper length are extracted as the key segment candidates. The final key segment is then selected from the candidate set according to a predefined criterion. With a low computation cost, the proposed method is feasible to process popular songs on line, or embedded into portable digital devices. An experiment is carried on a large music database with 259 popular songs. Compared with the key segments manually extracted, the performance of the proposed algorithm is rather satisfying.