Connectionist visualisation of tonal structure

摘要

Some forms of artificial neural network models develop representations that have a high visual information content. An example of this kind of network is the Kohonen Feature Map (KFM). This paper describes how a KFM can be used in a model that categorises memorised sequential patterns of notes into representations of key and degrees of a musical scale. These patterns are derived from abstractions of musical sounds identified with pitch and interval. Both key and degree are important musical structures in the cognitive organisation of tonality. The acquisition of tonal organisation for music is analogous to the acquisition of language. The representations developed within the KFM form a map that can be seen to correspond directly with the images used by musicians to represent key relations.