Tune Classification Using Multilevel Recursive Local Alignment Algorithms

C. Walshaw


This paper investigates several enhancements to two well-established local alignment algorithms in the context of their use for melodic similarity. It uses the annotated dataset from the well-known Meertens Tune Collection to provide a ground truth with which to evaluate the quality of the algorithms. In particular, recursive application of the alignment algorithms applied to a multilevel representation of the melodies is shown to be very effective for improving the accuracy of the classification of the tunes into families. However, the ideas should be equally applicable to music search and melodic matching.

Sat May 6 09:47:05 BST 2017