The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9823

Title:A Connectionis Model of Tension in chord Progressions
Date: 1998
Abstract:In this dissertation, the possibility of producing tonal music in real time while controlling the tension in it is studied by using a connectionist approach. An experiment to obtain tension data from human listeners is described and tested, and reliable results are obtained. A two-level structure and representation for producing music in real time, and the training of three back-propagation nets with music and tension examples to predict tensionk the following bar and the note distribution are described. A system composed of multilevel nets to produce music with tension content is described and tested. The results and behaviour of the nets and the program are analysed. A back-propagation net proves to be able to predict the tension, by responding to explainable harmonic patterns. The neural nets that produce bars nat notes learned important features of music but find undesirable steady-states easily; suggestions are made to avoid this. The idea is promising, and encourages further experimenting and research.

[Search These Pages] [DAI Home Page] [Comment]