Deep Learning Techniques for Music Generation - A Survey

This paper is a survey and an analysis of different ways of using deep learning (deep artificial neural networks) to generate musical content. We propose a methodology based on five dimensions for our analysis: - Objective - What musical content is to be generated? E.g., melody, polyphony, accompaniment and counterpoint - For what destination and for what use? To be performed by a human(s) or by a machine. - Representation - What are the concepts to be manipulated? E.g., waveform, spectrogram, note, chord, meter, and beat - What format is to be used? E.g., MIDI, piano roll and text - How will the representation be encoded? E.g., scalar, one-hot, and many-hot. - Architecture - What type of deep neural network is to be used? E.g., feedforward network, recurrent network, autoencoder, and generative adversarial networks. - Challenges - What are the limitations\index and open challenges? E.g., variability, interactivity and creativity. - Strategy - How do we model and control the process of generation? E.g., single-step feedforward, decoder feedforward, sampling and input manipulation. For each dimension, we conduct a comparative analysis of various models and techniques and we propose some tentative multidimensional typology. This typology is bottom-up, based on the analysis of many existing deep-learning based systems for music generation selected from the relevant literature. These systems are described in this survey/analysis and are used to exemplify the various choices of objective, representation, architecture, challenges and strategies. The final part of the paper includes some discussion and some prospects. This paper is a simplified (weak DRM) version of the following book: Jean-Pierre Briot, Ga\"etan Hadjeres and Fran\c{c}ois Pachet, Deep Learning Techniques for Music Generation, Computational Synthesis and Creative Systems, Springer Nature, 2019.