---
license: cc-by-3.0
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: relative_path
    dtype: string
  splits:
  - name: test
    num_bytes: 13646545685.208
    num_examples: 15292
  - name: validation
    num_bytes: 22378049262.984
    num_examples: 25468
  - name: train
    num_bytes: 269257423227.302
    num_examples: 150787
  download_size: 295850537553
  dataset_size: 305282018175.494
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
---

# MelodySim: Measuring Melody-aware Music Similarity for Plagiarism Detection

[Github](https://github.com/AMAAI-Lab/MelodySim) | [Model](https://huggingface.co/amaai-lab/MelodySim/tree/main) | [Paper](https://arxiv.org/abs/2505.20979)

The MelodySim dataset contains 1,710 valid synthesized pieces originated from Slakh2100 dataset, each containing 4 different versions (through various augmentation settings), with a total duration of 419 hours.

This dataset may help research in:
- Music similarity learning
- Music plagiarism detection

# Dataset Details
The MelodySim dataset contains three splits: train, validation and test. Each split contains multiple tracks. Each track folder contains the same song in 4 versions ("original", "version_0", "version_1", "version_2"), all of which are synthesized from the same midi file with [sf2](https://github.com/Rainbow-Dreamer/sf2_loader) in different settings. Checkout [MelodySim Paper](https://arxiv.org/abs/2505.20979) for details how the different versions are augmented. Each version contains multiple 10-second chunks named with their indices.

After downloading the dataset, [this dataloader](https://github.com/AMAAI-Lab/MelodySim/blob/main/data/dataloader.py) may help loading the dataset.


# Citation

If you find this work useful in your research, please cite:

```bibtex
@article{lu2025melodysim,
  title={Text2midi-InferAlign: Improving Symbolic Music Generation with Inference-Time Alignment},
  author={Tongyu Lu and Charlotta-Marlena Geist and Jan Melechovsky and Abhinaba Roy and Dorien Herremans},
  year={2025},
  journal={arXiv:2505.20979}
}
```