MUSDB18-HQ: Uncompressed Music Source Separation Dataset (150 Songs with Stems)
Rafii, Zafar and Liutkus, Antoine and Stöter, Fabian-Robert and Mimilakis, Stylianos Ioannis and Bittner, Rachel

@article{rafii2019musdb18hq,
  title={MUSDB18-HQ: Uncompressed Music Source Separation Dataset (150 Songs with Stems)},
  author={Rafii, Zafar and Liutkus, Antoine and St{"o}ter, Fabian-Robert and Mimilakis, Stylianos Ioannis and Bittner, Rachel},
  year={2019},
  url={https://zenodo.org/records/3338373},
  doi={10.5281/zenodo.3338373},
  keywords={music, audio, source separation, stems, SiSEC 2018},
  license={Non-commercial, educational use only (see License below)},
  abstract={**MUSDB18-HQ** is the uncompressed (WAV) version of MUSDB18: **150 full-length songs** in a range of styles, each with the stereo mixture and four separated stems. It is the standard reference dataset for music source separation and was the official dataset for the professionally produced music task of SiSEC 2018.

## What is in this torrent

A single file, `musdb18hq.zip` (22.7 GB), with two folders:

- `train`: 100 songs
- `test`: 50 songs

Each song folder holds `mixture.wav`, `drums.wav`, `bass.wav`, `other.wav` and `vocals.wav`. All signals are stereo, 44.1 kHz, uncompressed WAV. Supervised methods should train on `train` and evaluate on both subsets.

## Why it matters

The goal of source separation is to estimate one or more sources from a mixture, for example removing vocals for karaoke. MUSDB18-HQ gives a common benchmark so methods can be compared fairly.

## Links

- Original record: https://zenodo.org/records/3338373
- DOI: 10.5281/zenodo.3338373

## License

For educational and research purposes only. The material must not be used commercially without permission of the copyright holders. The tracks come from several sources:

- 100 tracks from the DSD100 set, derived from the Mixing Secrets Free Multitrack Download Library
- 46 tracks from MedleyDB (CC BY-NC-SA 4.0)
- 2 tracks from Native Instruments stems packs
- 2 tracks from The Easton Ellises, via the heise stems remix competition (CC BY-NC-SA 3.0)

## Citation

    Rafii, Z., Liutkus, A., Stöter, F.-R., Mimilakis, S. I., Bittner, R.
    MUSDB18-HQ: an uncompressed version of MUSDB18. Zenodo, 2019.
    https://doi.org/10.5281/zenodo.3338373

If you compare against SiSEC 2018 participants, also cite: Stöter, F.-R., Liutkus, A., Ito, N. The 2018 Signal Separation Evaluation Campaign. LVA/ICA 2018, pp. 293 to 305.
  }
}