MulTTiPop: A Multitrack Transcription Dataset for Pop Music

Nathan Pruyne, Benjamin Stoler, William Chen, Chien-yu Huang, Shinji Watanabe, Chris Donahue

🤗Download the Data: https://huggingface.co/datasets/gclef-cmu/multtipop

📄Read the Paper: ARXIV LINK

MulTTiPop is a dataset of segments of pop music, sourced from TheoryTab, and aligned multitrack MIDI transcriptions, sourced from the Lakh MIDI Dataset. MulTTiPop contains 572 total segments and approximately 3.5 hours of total data.

We release MulTTiPop as pairs of aligned multitrack MIDI files and associated metadata files with information about the audio. Specifically, the metadata contains YouTube video IDs and timestamps within the video for the audio that corresponds to the MIDI labels.

We make 2 key recommendations for the use of MulTTiPop:

  1. Researchers should only obtain the relevant segments of the original audio.
  2. Researchers should only use MulTTiPop for model evaluation, not training.

Splits

MulTTiPop is split in an approximately 3:7 ratio between dev and test splits. The dev split is intended for development, and the test split for model evaluation. As MulTTiPop is not intended for model training, we do not provide a train or validation split.

Citation

If you use MulTTiPop in your research, please cite our ArXiv paper:

bibtex for paper

License and Attribution

MulTTiPop is distributed with a CC-BY 4.0 License.

The aligned MIDI files provided in MulTTiPop are adapted from the Lakh MIDI Dataset by Colin Raffel, which is distributed with a CC-BY 4.0 License.

Select a song
YouTube
MIDI
MIDI:
🤗 Download MulTTiPop
If MIDI and audio sound out of sync, try refreshing the page