Modelling musical cognition with artificial neural networks

Abstract

As a highly abstract form of human activity, music is a challenging realm to study. During the last ten years, the connectionist paradigm has provided insights into many domains of human behaviour, including musical activity and experience. Artificial neural networks, or connectionist systems, can be characterized as strongly idealized models of networks formed by biological neurons: consisting of a bulk of simple interconnected processing units, they employ parallel distributed processing and are capable of learning and self-organizing. The present study focuses on aspects of musical cognition such as perceptual learning, self-organization, feature extraction, sequential processing, autoassociative recall, and short-term memory. More specifically, processes related to the classification and recognition of musical timbre and the learning and generation of melodies are modelled using artificial neural networks. The results support the view that the connectionist paradigm provides a plausible alternative for modelling the dynamics of certain music-related cognitive processes. Being inherently capable of generalizing, associating on the basis of content, and tolerating noisy or distorted input, artificial neural networks exhibit functions characteristic of the human way of perceiving, thinking, and acting.