Indexing Music Collections Through Graph Spectra

Content based music retrieval opens up large collections, both for the general public and music scholars. It basically enables the user to find (groups of) similar melodies, thus facilitating musicological research of many kinds. We present a graph spectral approach, new to the music retrieval field, in which melodies are represented as graphs, based on the intervals between the notes they are composed of. These graphs are then indexed into a database using their laplacian spectra as a feature vector. This laplacian spectrum is known to be very informative about the graph, and is therefore a good representative of the original melody. Consequently, range searching around the query spectrum returns similar melodies. We present an experimental evaluation of this approach, together with a comparison with two known retrieval techniques. On our test corpus, a subset of a well documented and annotated collection of Dutch folk songs, this evaluation demonstrates the effectiveness of the overall approach.