/HARD-DAFx2023

Primary LanguageJupyter NotebookGNU General Public License v2.0GPL-2.0

Self-Supervised Disentanglement of Harmonic and Rhythmic Features in Music Audio Signals

Source codes for the DAFx2023 conference paper "Self-Supervised Disentanglement of Harmonic and Rhythmic Features in Music Audio Signals".

How to train

  1. Store the audio files uncer a certain path,
  2. In configs/disentangle-2dvaegan-mono.yaml, modify model.init_args.dset_config.path to specify the audio path,
  3. Run the following command.
python RunCLI.py fit --config configs/disentangle-2dvaegan-mono.yaml

Inference

An example code of audio generation is shown in Inference.ipynb.