softcatala/wikimedia-common-audio-catalan

Wikimedia Commons Audio — Catalan

A collection of Catalan-language audio recordings harvested from Wikimedia Commons, together with the per-file description and license metadata published on Commons.

Dataset summary

• Language: Catalan (ca)
• Files: 512 audio recordings (.ogg and .mp3)
• Total duration: ~30.7 hours
• Source: Wikimedia Commons (files categorised as Catalan-language audio)
• Modality: audio + text metadata

Recordings include poetry readings, interviews, radio programme excerpts, voice samples, dialectal recordings, spoken essays, and other spoken-word content contributed to Wikimedia Commons by its community.

Dataset structure

The dataset follows the Hugging Face audiofolder layout: a single metadata.csv file at the repository root pointing at audio files stored under catalan/.

.
├── README.md
├── metadata.csv
└── catalan/
    ├── <file>.ogg
    ├── <file>.mp3
    └── ...

Fields

Column	Type	Description
file_name	string	Relative path to the audio file (e.g. catalan/<file>.ogg).
description	string	Description of the recording as taken from Wikimedia Commons (in Catalan).
license	string	The license declared on the Commons file page for the recording.
duration	float	Duration of the audio file in seconds.

Splits

The dataset ships as a single train split. Users are free to derive their own train/dev/test partitions.

Loading

from datasets import load_dataset

ds = load_dataset("<user-or-org>/<repo-name>")
print(ds["train"][0])

Because each row carries its own license field, downstream users should filter by license before use if a specific license is required, for example:

cc0 = ds["train"].filter(lambda r: r["license"].lower() in {"cc-zero", "cc-0"})

Data collection

Audio files were located by scanning the Wikimedia Commons dump for Catalan-language audio pages, downloading the associated media, and extracting the human-authored description and license metadata directly from the Commons file pages. The processing scripts are available in the Softcatala/wikimedia-common-audio repository.

No new transcription or annotation was produced for this release: the description field reflects what contributors wrote on Wikimedia Commons, not a verbatim transcript of the audio.

Licensing

This dataset is a compilation of files under multiple free licenses. The license that applies to each individual audio file is recorded in the license column and must be respected on a per-file basis. The distribution of licenses across the collection is approximately:

License	Files
CC0 / Public Domain (cc-zero, cc-0)	278
CC BY-SA 4.0	223
CC BY-SA 3.0	8
GFDL	1
PD-self	1
Other / unspecified	1

• CC0 / Public Domain: no restrictions.
• CC BY-SA: you must give appropriate credit to the original contributor (see the corresponding file page on Wikimedia Commons) and distribute any derivatives under the same license.
• GFDL: subject to the terms of the GNU Free Documentation License.

Attribution should point back to the file's page on Wikimedia Commons, which lists the original author. The compilation itself does not add any additional restrictions beyond those of the individual files.

For the general Wikimedia Commons licensing policy, see https://commons.wikimedia.org/wiki/Commons:Licensing.

Considerations for using the data

• No verified transcripts. The description field is a Commons caption, not an aligned transcript. Do not treat it as ground-truth text for ASR training without further processing.
• Heterogeneous recording conditions. Files come from many contributors and were recorded on a wide variety of equipment and in varied environments; audio quality, sample rates, and channel counts are not uniform.
• Speaker demographics are unknown. No speaker identity, age, gender, regional variety, or consent-for-ML metadata is systematically attached, beyond what individual Commons pages may mention.
• Personal data. Some recordings feature named public figures (writers, journalists, politicians, etc.) speaking in a public capacity. Downstream users are responsible for complying with applicable privacy and data-protection laws when redistributing or training on this material.

Citation

If you use this dataset, please cite Wikimedia Commons as the upstream source and credit the individual contributors listed on each file page.

@misc{wikimedia_commons_audio_catalan,
  title  = {Wikimedia Commons Audio — Catalan},
  author = {Wikimedia Commons contributors},
  howpublished = {\url{https://commons.wikimedia.org/}},
  note   = {Compiled dataset of Catalan-language audio from Wikimedia Commons}
}