Automatic harmonic description of musical signals using schema-based chord decomposition

This paper presents a model for the harmonic description of musical signals using schema-based chord decomposition. The model is based (i) on a bottom-up processing of musical signals into auditory images of different kinds and (ii) on a schema-based top-down processing of the images which produces chord decomposition. The schema is carried by a neural network and its content is derived from the self-organization of a pre-selected set of chords in all keys. The system architecture of this model is described and its performance is analyzed in detail through several excerpts of piano pieces. An evaluation procedure has been developed to compare the results with the ones derived using a music theoretical approach to chord decomposition.