We compare two methods of measuring melodic similarity for symbolically represented polyphonic music. Both exploit advantages of transportation distances such as continuity and partial matching in the pitch dimension. By segmenting queries and database documents, one of them also offers partial matching in the time dimension. This method can find short queries in long database documents and is more robust against pitch and tempo fluctuations in the queries or database documents than the method that uses transportation distances alone. We compare the use of transportation distances with and without segmentation for the RISM A/II collection and find that segmentation improves recall and precision. With everything else being equal, the segmented search found 80 out of 114 relevant documents, while the method relying solely on transportation distances found only 60.
Rainer Typke, Frans Wiering, Remco C. Veltkamp