A Hybrid Neural-Markov Approach for Learning to Compose Music by Example

Abstract

In this paper we introduce a hybrid approach to autonomous music composition by example. Our approach utilizes pattern recognition, Markov chains, and neural networks. We first extract patterns from existing musical training sequences, and then construct a Markov chain based on these patterns with each state corresponding to a pattern. We then use a neural network to learn which shifts of pitch and duration are allowed for each pattern in the training sequences. Using this hybrid model, we compose novel musical sequences.