Indexing musical pieces using their major repetition

With the growing presence of large collections of musical content, methods for facilitating efficient browsing and fast comparisons of audio pieces become more and more useful. Notably, methods that isolate relevant parts in audio pieces give an insight of the musical content and can be used to improve similarity evaluation systems. In this context, we propose an indexing method that allows retrieving in audio signals particular parts, namely a major repetition. We use harmonic representations together with string matching techniques to strictly define and isolate such segments. Experiments on state-of-the-art structural datasets show a strong correlation between the retrieved parts and the perceived structure of pieces.