espnet/wikitongues

The WikiTongues speech corpus is a collection of conversational audio across 700+ languages. It can be used for spoken language modelling or speech representation learning. This dataset includes the raw unsegmented audio in a 16kHz single channel format. Each clip is usually 2-10 minutes long, and contains one or more speakers conversing in their language(s). Sometimes, a speaker may switch languages within a single clip. The total dataset size is around 70 hours.

The current version of the dataset does not include labels for the language(s) being spoken in each clip. This information will be included in an update in the near future

This dataset was crawled from the WikiTongues project, which collected the original recordings. We use this corpus to train XEUS, a multilingual speech encoder for 4000+ languages. For more details about the dataset and its usage, please refer to our paper or project page.

License and Acknowledgement

WikiTongues is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 license.

If you use this dataset, we ask that you cite our paper:

@misc{chen2024robustspeechrepresentationlearning,
      title={Towards Robust Speech Representation Learning for Thousands of Languages}, 
      author={William Chen and Wangyou Zhang and Yifan Peng and Xinjian Li and Jinchuan Tian and Jiatong Shi and Xuankai Chang and Soumi Maiti and Karen Livescu and Shinji Watanabe},
      year={2024},
      eprint={2407.00837},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2407.00837}, 
}

And credit the original creators of the audio.