Melodic similarity as a determinant of melody structure

Abstract

This paper presents an approach to the analysis of melodic similarity as a determinant of melody structure, which has been applied successfully as a computational system of analysis for the prediction of segmental structure in a large sample of monophonic melodies of diverse cultural origin. (Ahlbäck 2004) Although the impact of similarity in the determination of segmental structure in music is generally acknowledged, methods based on similarity have been criticized with regards to the difficulty of formalizing criteria and threshold values for structurally significant similarity (e.g. Cook 1987). It is herein maintained that, since similarity is a fundamentally relative concept and categorization of similarity and difference relates to a given context, segmentation based on similarity requires a context-sensitive parametric measure of similarity. The proposed method of analysis is based on common psychological principles such as gestalt psychological principles, human perceptual and cognitive limitations regarding temporal frames of attention, simultaneous category handling and cognition of temporal proportions. The core hypothesis of the model is that melodic segmentation at a structural level is primarily established by similarity, in particular repetition of melodic content, and dissimilarity, in particular discontinuity of melodic processes. It is discussed how the influence of melodic similarity on segment structure is dependent upon general features of musical structure, such as metrical structure. This notion is supported by the results of an experiment, which indicates that listeners do not make use of repetitions of pitch sequences for melodic segmentation when these are in conflict with the perceived general beat structure. The influence of metrical context is handled within the method, as well as the means of allowing for different levels of similarity through a categorization of different types and degree of similarity. This is illustrated by an example analysis of a melody in which the segmental structure is determined by melodic similarity of different types, the result of which is evaluated by a listener test. The result supports the notion that structurally significant similarity is relative and contextual and indicates that this is possible to model formally within a rule-based, style-independent method of analysis.