Discovering Musical Patterns through Perceptive Heuristics

This paper defends the view that the intricate difficulties challenging the emerging domain of Musical Pattern Discovery, which is dedicated to the automation of motivic analysis, will be overcome only through a thorough taking into account of the specificity of music as a perceptive object. Actual musical patterns, although constantly transformed, are nevertheless perceived by the listener as musical identities. Such dynamical properties of human perception, not reducible to geometrical models, will only be explained with the notions of contexts and expectations. This paper sketches the general principles of a new approach that attempts to build such a general perceptual system. On a sub-cognitive level, patterns are discovered through the detection, by an associative memory, of local similarities. On a cognitive level, patterns are managed by a general logical framework that avoids irrelevant inferences and combinatorial explosion. In this way, actual musical patterns that convey musical significance are discovered. This approach, offering promising results, is a first step toward a complete system of automated music analysis and an explicit modeling of basic mechanisms for music understanding.