A study on content-based music classification

Abstract

Content-based music recognition can play an important role in human cognition research and multimedia applications. In this paper, we present a study on content-based music classification using short-time analysis techniques together with pattern recognition techniques to distinguish between five music styles. A database of total 1027 audio signals (99 piano, 204 symphony, 304 popular song, 242 Beijing opera, and 178 Chinese comic dialogues) is used for the experiments, which is much larger than the previous works. A comparative evaluation between different short-time features in terms of their classification ability, as well as between different classifiers is carried out on the database. The results show that harmonious degree is the most effective feature and the BPNNC is the best classifier. Some interesting results about different music styles are also reported.