An integrative computational modelling of music structure apprehension

An objectivization of music analysis requires a detailed formalization of the underlying principles and methods. The formalization of the most elementary structural processes is hindered by the complexity of music, both in terms of profusions of entities (such as notes) and of tight interactions between a large number of dimensions. Computational modeling would enable systematic and exhaustive tests on sizeable pieces of music, yet current researches cover particular musical dimensions with limited success. The aim of this research is to conceive a computational modeling of music analysis encompassing the main core components and focusing on their interdependencies. The system should be as simple as possible, while producing relevant structural analyses on a large variety of music. This paper describes the general principles of a computational framework for music analysis currently under construction. The complex system of music structure is described as an interaction between modules focusing on formal operations that are conceived as general as possible. Each module addresses a core aspect of music analysis and offers some innovative breakthrough compared to the state of the art. In order to overcome the limitations of local segmentation, we propose an alternative paradigm based on hierarchical local grouping. Ornamentation reduction, and more generally the inference of deeper structures from the musical surface is carried out by selecting key elements from each local grouping structure, offering hence an implementation of “Time-Span Reduction”. Parallelism, i.e. sequential repetition, considered here as an essential aspect of music analysis, is applied to the search for motives, for mode-related patterns as well as metrical analysis. A new approach for modal analysis is based on a comparison of the local context (defined by the current and recent notes, and taking into account ornaments reduction) with all possible modes and key scales. Structural information inferred in each module is fed back into the other modules. Besides the practical application to automated music analysis, the computational model, by virtue of its generality, extensiveness and operationality, is suggested as a blueprint for the establishment of cognitively validated model of music structure apprehension. Available as a Matlab module, it can be used for practical musicological uses.