Using Pitch Stability Among a Group of Aligned Query Melodies to Retrieve Unidentified Variant Melodies

Melody identification is an important task in folk song variation research. In this paper we develop methods and tools that support researchers in finding melodies in a database that belong to the same variant group as a set of given melodies. The basic approach is to derive from the pitches of the known variants per onset a weighted pitch distribution, which quantifies pitch stability. We allow for partial matching and AND and OR queries. Technically we do so by defining a distance measure between weighted pitch distribution sequences. It is based on two applications of the Earth Mover’s Distance, which is a distribution distance. We set up a distance framework and discuss musically meaningful parameterizations for two tasks: a) Study the inner-group distances between the group as a whole and single members of the group. b) Use the group’s weighted pitch distribution sequence to query for variant melodies. The first experimental results seem very promising: a) The inner-group distances correlate to expert assigned subgroups. b) For variant retrieval our method works better than last year’s MIREX winner.