Folk music style modelling by recurrent neural networks with long short term memory units

Abstract

We demonstrate two generative models created by training\na recurrent neural network (RNN) with three hidden\nlayers of long short-term memory (LSTM) units. This extends\npast work in numerous directions, including training\ndeeper models with nearly 24,000 high-level transcriptions\nof folk tunes. We discuss our on-going work.