In this paper we proposed a music matching method, called the RLCS(rough longest common subsequence) method. It is an impoved version of the LCS to avoid some problems occurring in global alignment matching. First a rough equality for two notes is defined for constructing the RLCS of two music fragments. The length of the RLCS of two music sequences defined in this work is a real number, called a weighted length. It is evaluated according to degree of similarity of every pair of matched notes from the two sequences. This method takes into account both the width-across-query(WAQ) and the width-across-reference(WAR) and combines them with the weighted length of the corresponding RLCS to define a score measurement for the RLCS. The measurement associated with WAQ and WAR enables the proposed method to tolerate dense errors. A dynamic programming algorithm is presented for simultaneously calculating the weighted length of RLCS, the WAQ, the WAR, and the score to determine the RLCS. As a result, the proposed method can perform the matching in a better and simplermanner. In order to speed up the matching process, we use the filter algorithm proposed by Tarhio and Ukkonen  to filter the reference and discard most off the reference areas that do not match. We applied the proposed algorithm to content-based music retrieval. The experimental results showed that with our proposed algorithm the retrieval system provides a higher retrieval rate than that with the local alignment method proposed by Suyoto et al. . The use of filtering algorithm has been shown to greatly reduce the computation time for exact matching and for approximate matching with a low error tolerance.
Journal of Information Science and Engineering 27(1), pp.95-110