parler-tts/mls_eng_10k

Dataset Summary

This is a 10K hours subset of English version of the Multilingual LibriSpeech (MLS) dataset. The data archives were restructured from the original ones from OpenSLR to make it easier to stream.

MLS dataset is a large multilingual corpus suitable for speech research. The dataset is derived from read audiobooks from LibriVox and consists of 8 languages - English, German, Dutch, Spanish, French, Italian, Portuguese, Polish. It includes about 44.5K hours of English and a total of about 6K hours for other languages.

This dataset card includes the 10K hours of English. Refers to this dataset card for the other languages.

Licensing Information

Public Domain, Creative Commons Attribution 4.0 International Public License (CC-BY-4.0)

Citation Information

@article{Pratap2020MLSAL,
  title={MLS: A Large-Scale Multilingual Dataset for Speech Research},
  author={Vineel Pratap and Qiantong Xu and Anuroop Sriram and Gabriel Synnaeve and Ronan Collobert},
  journal={ArXiv},
  year={2020},
  volume={abs/2012.03411}
}