Musical Pattern Extraction for Melodic Segmentation

Despite the fact that musical parallelism is considered as an important factor for musical segmentation, there have been very few attempts to describe systematically how exactly it affects grouping processes. The main problem is that musical parallelism itself is very difficult to formalise. In this paper a computational model will be presented that extracts melodic patterns from a given melodic surface. Following the assumption that the beginning and ending points of 'significant' repeating musical patterns influence the segmentation of a musical surface, the discovered patterns are used as a means to determine probable segmentation points of the melody. ‘Significant’ patterns are defined primarily in terms of frequency of occurrence and length of pattern. The special status of non-overlapping immediately-repeating patterns is examined. All the discovered patterns merge into a single ‘pattern’ segmentation profile that signifies points in the surface that are most likely to be perceived as points of segmentation. The effectiveness of the algorithm is tested against a series of musical surfaces illustrating both strengths and weaknesses of the approach.