Free Music Archive
Collection
ISMIR's 2017 FMA Dataset, Optimized for 🤗 Datasets / 🥐 Croissant, with Clear Licensing
•
4 items
•
Updated
•
3
The dataset viewer is not available for this split.
Rows from parquet row groups are too big to be read: 467.72 MiB (max=286.10 MiB)
Error code: TooBigContentError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
FMA: A Dataset for Music Analysis
Michaël Defferrard, Kirell Benzi, Pierre Vandergheynst, Xavier Bresson.
International Society for Music Information Retrieval Conference (ISMIR), 2017.
We introduce the Free Music Archive (FMA), an open and easily accessible dataset suitable for evaluating several tasks in MIR, a field concerned with browsing, searching, and organizing large music collections. The community's growing interest in feature and end-to-end learning is however restrained by the limited availability of large audio datasets. The FMA aims to overcome this hurdle by providing 917 GiB and 343 days of Creative Commons-licensed audio from 106,574 tracks from 16,341 artists and 14,854 albums, arranged in a hierarchical taxonomy of 161 genres. It provides full-length and high-quality audio, pre-computed features, together with track- and user-level metadata, tags, and free-form text such as biographies. We here describe the dataset and how it was created, propose a train/validation/test split and three subsets, discuss some suitable MIR tasks, and evaluate some baselines for genre recognition. Code, data, and usage examples are available at https://github.com/mdeff/fma.
Paper: arXiv:1612.01840 - latex and reviews
Slides: doi:10.5281/zenodo.1066119
Poster: doi:10.5281/zenodo.1035847
This Pack
This is the full dataset, comprising a total of 106,199 clips of untrimmed length over 16 unbalanced genres totaling 8,104 hours of audio.
Packed as Parquet files, this dataset is 593 GB in size, roughly a 34% size saving over the original ZIP file.
Repack Notes
- 173 files were unreadable by
libsndfile / libmpg123, these were removed. - 202 files had licenses that were unclear on whether or not they permitted redistribution, or the full license text was unavailable. These were removed.
- Many of the remaining files had mixed or inconsistent encoding. To homogenize the dataset, all audio was re-encoded using
libmpg123.
License
- The FMA codebase is released under The MIT License.
- The FMA metadata is released under CC-BY 4.0.
- The individual files are released under various Creative Commons family licenses, with a small amount of additional licenses. Each file has its license attached and important details of the license enumerated. To make it easy to use for developers and trainers, a configuration is available to limit only to commercially-usable data.
Please refer to any of the following URLs for additional details.
Total Duration by License
| License | Total Duration (Percentage) |
|---|---|
| CC-BY-NC-SA 3.0 | 2768.3 hours (34.16%) |
| CC-BY-NC-ND 3.0 | 2296.4 hours (28.34%) |
| CC-BY-NC-ND 4.0 | 1018.4 hours (12.57%) |
| CC-BY-NC-SA 4.0 | 533.2 hours (6.58%) |
| CC-BY 4.0 | 377.0 hours (4.65%) |
| CC-BY-NC 3.0 | 288.9 hours (3.56%) |
| CC-BY-NC 4.0 | 232.6 hours (2.87%) |
| CC-BY 3.0 | 106.9 hours (1.32%) |
| CC-BY-SA 4.0 | 99.4 hours (1.23%) |
| CC-BY-SA 3.0 | 79.7 hours (0.98%) |
| CC-BY-NC-SA 2.0 | 65.1 hours (0.80%) |
| CC-BY-NC-ND 2.0 | 56.2 hours (0.69%) |
| CC-BY-ND 3.0 | 36.8 hours (0.45%) |
| CC-BY-ND 4.0 | 25.0 hours (0.31%) |
| CC-BY-NC-ND 2.5 | 24.2 hours (0.30%) |
| FMA Sound Recording Common Law | 19.9 hours (0.25%) |
| CC-BY-NC-SA 2.5 | 18.0 hours (0.22%) |
| CC-BY-NC 2.5 | 13.3 hours (0.16%) |
| CC0 1.0 | 10.5 hours (0.13%) |
| CC-BY 1.0 | 10.4 hours (0.13%) |
| Free Music Philosophy (FMP) | 4.4 hours (0.05%) |
| Free Art License | 2.7 hours (0.03%) |
| CC-BY 2.0 | 2.5 hours (0.03%) |
| CC-BY-NC 2.1 | 2.4 hours (0.03%) |
| CC-BY-NC-SA 2.1 | 2.3 hours (0.03%) |
| CC-BY-SA 2.0 | 1.9 hours (0.02%) |
| CC-BY-NC 2.0 | 1.6 hours (0.02%) |
| CC-BY-ND 2.5 | 1.6 hours (0.02%) |
| CC-NC-Sampling+ 1.0 | 1.4 hours (0.02%) |
| CC-BY-NC-ND 2.1 | 65.0 minutes (0.01%) |
| CC-Sampling+ 1.0 | 53.9 minutes (0.01%) |
| CC-BY-SA 2.5 | 31.8 minutes (0.01%) |
| CC-BY-ND 2.0 | 29.7 minutes (0.01%) |
| CC-BY 2.5 | 11.2 minutes (0.00%) |
Citations
@inproceedings{fma_dataset,
title = {{FMA}: A Dataset for Music Analysis},
author = {Defferrard, Micha\"el and Benzi, Kirell and Vandergheynst, Pierre and Bresson, Xavier},
booktitle = {18th International Society for Music Information Retrieval Conference (ISMIR)},
year = {2017},
archiveprefix = {arXiv},
eprint = {1612.01840},
url = {https://arxiv.org/abs/1612.01840},
}
@inproceedings{fma_challenge,
title = {Learning to Recognize Musical Genre from Audio},
subtitle = {Challenge Overview},
author = {Defferrard, Micha\"el and Mohanty, Sharada P. and Carroll, Sean F. and Salath\'e, Marcel},
booktitle = {The 2018 Web Conference Companion},
year = {2018},
publisher = {ACM Press},
isbn = {9781450356404},
doi = {10.1145/3184558.3192310},
archiveprefix = {arXiv},
eprint = {1803.05337},
url = {https://arxiv.org/abs/1803.05337},
}
- Downloads last month
- 361
Size of the auto-converted Parquet files (First 5GB):
5 GB
Number of rows (First 5GB):
1,100
Estimated number of rows:
106,890