README.md

---
license: cc-by-nc-nd-4.0
task_categories:
- automatic-speech-recognition
language:
- fr
size_categories:
- 100K<n<1M
---
The french subset of the dataset [Multilingual TEDx](https://www.openslr.org/100). The data uploaded to HF corresponds to the directory fr-fr

Configs:
- single_samples (default): all samples taken separately
- max=30s: combine consecutive samples for a period shorter than 30 seconds
- max=10s: combine consecutive samples for a period shorter than 10 seconds
- max: combine all the samples of a TEDx talk

```

dependencies:
 - ffmpeg: apt install ffmpeg
 - ffmpeg-python: pip install ffmpeg-python

@inproceedings{salesky2021mtedx,
  title={Multilingual TEDx Corpus for Speech Recognition and Translation},
  author={Elizabeth Salesky and Matthew Wiesner and Jacob Bremerman and Roldano Cattoni and Matteo Negri and Marco Turchi and Douglas W. Oard and Matt Post},
  booktitle={Proceedings of Interspeech},
  year={2021},
}
```