TalBarami/msd_dmelodies_wav

MSD dMelodies-WAV Dataset Attribution

The Multi-factor Sequential Disentanglement benchmark includes the dMelodies-WAV dataset, a synthetically generated collection of 48,000-sample audio waveforms labeled with one static factor (instrument) and five dynamic musical attributes. Extending the symbolic dMelodies benchmark into the raw audio domain, dMelodies-WAV was created by synthesizing a subset of dMelodies using the MIDI-DDSP neural audio synthesis model, producing realistic audio across four instruments. This dataset provides both global factors (instrument, tonic, mode) and local factors (rhythm, arpeggiation), making it a strong disentanglement benchmark for studying global-local and hierarchical factor models in raw waveform music.

• Original repositories:

  https://github.com/ashispati/dmelodies_dataset

  https://github.com/magenta/midi-ddsp

• Reference papers:
  A. Pati, S. Gururani, A. Lerch.
  dMelodies: A Music Dataset for Disentanglement Learning, ISMIR 2020.
  https://arxiv.org/abs/2007.15067

  @inproceedings{pati2020dmelodies,
    title={dMelodies: A Music Dataset for Disentanglement Learning},
    author={Pati, Ashis and Gururani, Siddharth and Lerch, Alexander},
    booktitle={21st International Society for Music Information Retrieval Conference (ISMIR)},
    year={2020},
    address={Montréal, Canada}
  }
  

  Y. Wu, E. Manilow, Y. Deng, R. Swavely, K. Kastner, T. Cooijmans, A. Courville, C. Auang, J, Engel MIDI-DDSP: Detailed Control of Musical Performance via Hierarchical Modeling, ISMIR 2021.
  https://arxiv.org/abs/2112.09312

  @inproceedings{
  wu2022mididdsp,
  title={{MIDI}-{DDSP}: Detailed Control of Musical Performance via Hierarchical Modeling},
  author={Yusong Wu and Ethan Manilow and Yi Deng and Rigel Swavely and Kyle Kastner and Tim Cooijmans and Aaron Courville and Cheng-Zhi Anna Huang and Jesse Engel},
  booktitle={International Conference on Learning Representations},
  year={2022},
  url={https://openreview.net/forum?id=UseMOjWENv}
  }
  

⚠ Note: The dMelodies-WAV dataset is provided for non-commercial research purposes. Please cite the above when using this dataset in your work.