Music composition is a suitable domain for applying the technique of neural networks where the number of ways to form a melody is infinite and the outcomes from the network model are unpredictable. Genetic algorithm is applied on top of the neural networks to maximize the chance to find good melodies in the global solution space. Here we introduce an approach to constrain the evolutionary process so that the outcomes are fit according to the constraints. Melodic constraints on tonality and rhythm are given to determine the fitness of the trial melodies in each generation. The best melody evolved over a period of time is analyzed for its performance. We observe that the network learns to use primitve rules on tonality and rhythm to come up melodies that embed certain characteristics, i.e. transposition.

PDF

Technical Report HR-00-02, Department of Computer Sciences, The University of Texas at Austin.

@techreport{chen:ugthesis00, title={Automatic Music Composition using Genetic Algorithm and Neural Networks: A Constrained Evolution Approach}, author={Chun-Chi Chen}, number={HR-00-02}, school={Department of Computer Sciences, The University of Texas at Austin}, institution={Department of Computer Sciences, The University of Texas at Austin}, type={Undergraduate Honors Thesis}, url="http://www.cs.utexas.edu/users/ai-lab?chen:ugthesis00", year={2000} }