Cover song identification using an enhanced chroma over a binary classifier based similarity measurement framework

Identifying all covers/versions of a query song from music collection is a challenging task since there exists much variance of multiple aspects, such as timbre, tempo, key, structure, among covers. In this paper we propose a cover song identification algorithm, about which there are two innovations. The first, we propose a method for extracting an enhanced chromagram which retains the harmonic partials of music and holds invariance of volume; the second, based on aforementioned chromagram, a similarity measurement framework where any binary classifier can be applied is schemed. As a case, we apply Bayes classifier to the framework, and experiments indicate the proposed algorithm is able to provide competitive retrieval accuracy.