Interest in the so-called “computer music” (i.e. not only Electronic Music or music composed on computer, but every kind of music that can be represented in electronic formats) are not only in the direction of music performance and reproduction. Even before the great diffusion of Personal Computers, Winograd discussed a “Linguistic Analysis” of Tonal Harmony for Computers, indicating a grammatical way to music description, that it is one of the most typical and abstract approach to music. However, this introduces some problems, concerning a principled generative view on musical elements and categories, but derived from an a priori perspective.
Nowadays, the evolution of Artificial Intelligence studies suggests to abandon a static grammar-based definition and to concentrate on the representational aspects that should not be postulated a priori, but acquired by experience. The aim of this reasearch is to explore the possibility of applying a machine learning perspective on the acquisition of prototypical musical properties characterizing music classes (like genres) from living examples, of instances of reference classes. Such an inductive approach has benefits on a methodological ground and also supports a variety of useful research lines, looking at the aggregation of existing repositories and the analysis of virtual communities (e.g. customers, buyers, of similar music material), explaining models for technical and social aspects of genre development and definition.
- (2007): Audio Feature Engineering for Automatic Music Genre Classification. In: Proceedings of RIAO, , Pittsburgh, USA, 2007.
- (2005): Extracting music features with midxlog. In: Dins Proc. ISMIR, 2005.