By Peter van Kranenburg, Anja Volk, and Frans Wiering
The recurrence of characteristic motifs plays an important role in the identification of a folk song melody as member of a tune family. Based on a unique data set with expert annotations of motif occurrences in a collection of Dutch folk song melodies, we define 15 abstract motif classes. Taking a computational approach, we evaluate to what extent these 15 motif classes contribute to automatic identification of folk songs. We define various similarity measures for melodies represented as sequences of motif occurrences. In a retrieval experiment, alignment measures appear the most successful. The results are additionally improved by taking into account the phrase position of motif occurrences. These insights motivate future research to improve automatic motif detection and retrieval performance, and to determine similarity between melodies on the basis of motifs.
This paper will be presented at the 12th International Conference on Music Perception and Cognition, July 23-28, 2012 in Thessaloniki, Greece.