Source: Mozilla Common Voice 22.0
Processing: Sidon denoising (sarulab-speech/sidon-v0.1) with 21 s chunks and 48 kHz reconstruction
Format: WebDataset shards (.tar.gz)
Manifest: paths.yaml enumerates every shard path for Hugging Face–style loading
License: Original Common Voice license (CC0 1.0)

Languages

137 language folders are available; the directory names mirror the Common Voice language identifiers.

ab af am ar as ast az ba bas be bg bn br ca ckb cnh cs cv cy da dav de dv dyu el en eo es et eu fa fi fr fy-NL ga-IE gl gn ha he hi hsb ht hu hy-AM ia id ig is it ja ka kab kk kln kmr ko ky lg lij lo lt ltg luo lv mdf mhr mk ml mn mr mrj mt myv nan-tw nb-NO ne-NP nhi nl nn-NO nr nso oc or os pa-IN pl ps pt quy rm-sursilv rm-vallader ro ru rup rw sah sat sc sd sk skr sl sq sr st sv-SE sw ta te tg th ti tig tk tn tok tr ts tt tw ug uk ur uz ve vi vot xh yi yo yue zgh zh-CN zh-HK zh-TW zu zza

Each language directory contains Sidon-processed shards for the Common Voice splits that exist for that language (train, validation, test, other, and invalidated).

Approximate shard counts across all languages:

train: 1529 shards
validation: 209 shards
test: 219 shards
other: 1384 shards
invalidated: 418 shards

Shard Contents

Every shard stores up to 5,000 utterances. Within a shard you will find records structured as:

123456.flac           # 48 kHz mono audio after Sidon restoration
123456.metadata.json  # Original Common Voice metadata (speaker info, split ids, transcript, etc.)

Usage Examples

Hugging Face Datasets (paths manifest)

import datasets
from IPython.display import Audio
from huggingface_hub import hf_hub_download
import yaml
token="YOUR_HF_ACCESS_TOKEN"
base_url = "https://huggingface.co/datasets/sarulab-speech/commonvoice22_sidon/resolve/main/"
language = 'en'
split = 'train'
data_file_path = hf_hub_download(repo_id="sarulab-speech/commonvoice22_sidon", repo_type="dataset", filename="paths.yaml",token=token)
paths = yaml.load(open(data_file_path, "r"), Loader=yaml.FullLoader)

ds = datasets.load_dataset("webdataset", data_files=[base_url + p for p in paths[language][split]],streaming=True,token=token)['train']

sample = next(iter(ds))
audio = sample['flac']
print(sample['metadata.json'])
Audio(audio['array'], rate=audio['sampling_rate'])

Notes

Sidon resamples chunks to 16 kHz internally and writes 48 kHz FLAC output.
Failed or skipped records are omitted, so counts can differ slightly from the raw Common Voice release.

Citation

If you use this dataset, please cite Sidon and Mozilla Common Voice.

@misc{nakata2025sidonfastrobustopensource,
      title={Sidon: Fast and Robust Open-Source Multilingual Speech Restoration for Large-scale Dataset Cleansing},
      author={Nakata, Wataru and Saito, Yuki and Ueda, Yota and Saruwatari, Hiroshi},
      year={2025},
      eprint={2509.17052},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2509.17052}
}

@article{ardila2020common,
  author    = {Ardila, Rosana and Branson, Megan and Davis, Kelly and Kohler, Michael and Meyer, Gabriela and Henretty, Reuben and Morais, Michael and Saunders, Lindsay and Tyers, Francis and Weber, Gregor},
  title     = {{Common Voice}: A Massively-Multilingual Speech Corpus},
  journal   = {Proceedings of the 12th Language Resources and Evaluation Conference (LREC)},
  year      = {2020}
}

License

Derived from Mozilla Common Voice, this dataset remains available under CC0 1.0. Please respect any additional requirements imposed by downstream usage (e.g., speaker consent in redistribution policies).

Acknowledgements

Mozilla and the Common Voice contributors for collecting and releasing the original data.

Downloads last month: 465