Deep Jammer: A Music Generation Model

Music generation remains an attractive and interesting application of machine learning since it is typically characterized by human ingenuity and creativity. Moreover, given the high-dimensionality of time-series data, it is difficult to construct a model that has the representational power necessary to capture the timeand note-invariant patterns throughout a musical piece. In this paper, we describe our implementation of a classical music generation model heavily influenced by previous work that uses deep neural networks, particularly Long-Short Term Memory networks (LSTMs), to capture the note and temporal patterns within a large dataset of classical pieces. We then use methods in transfer learning to train our classical music generation model on a small dataset of jazz pieces. Finally, we report and analyze the results of our experiments by comparing it to an existing music generation model that uses Markov