FIGARO: Generating Symbolic Music with Fine-Grained Artistic Control

  • 2022-01-26 13:51:19
  • Dimitri von Rütte, Luca Biggio, Yannic Kilcher, Thomas Hoffman
  • 63

Abstract

Generating music with deep neural networks has been an area of activeresearch in recent years. While the quality of generated samples has beensteadily increasing, most methods are only able to exert minimal control overthe generated sequence, if any. We propose the self-supervised\emph{description-to-sequence} task, which allows for fine-grained controllablegeneration on a global level by extracting high-level features about the targetsequence and learning the conditional distribution of sequences given thecorresponding high-level description in a sequence-to-sequence modelling setup.We train FIGARO (FIne-grained music Generation via Attention-based, RObustcontrol) by applying \emph{description-to-sequence} modelling to symbolicmusic. By combining learned high level features with domain knowledge, whichacts as a strong inductive bias, the model achieves state-of-the-art results incontrollable symbolic music generation and generalizes well beyond the trainingdistribution.

 

Quick Read (beta)

loading the full paper ...