metadata
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. The data uploaded to HF corresponds to the directory fr-fr. The audio files are automatically resampled to 16 kHz.
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 (only needed for much faster audio decoding):
- ffmpeg: apt install ffmpeg
- ffmpeg-python: pip install ffmpeg-python
Sample
{'file': '0u7tTptBo9I-0', 'audio': {'path': None, 'array': array([ 3.05175781e-05, 6.10351562e-05, 9.15527344e-05, ...,
-2.44140625e-04, -3.35693359e-04, -2.74658203e-04]), 'sampling_rate': 16000}, 'sentence': "Bonsoir ! Notre planète est recouverte à 70 % d'océan, et pourtant, étrangement, on a choisi de l'appeler « la Terre ». Le poète Heathcote Williams a une vision bien plus objective et moins anthropocentrique, quand il dit que « Vue de l'espace, la planète est bleue. Vue de l'espace, elle est le territoire, non pas des hommes, mais des baleines ». Et pourtant, on vient tous de l'océan. ", 'speaker_id': '0u7tTptBo9I', 'start_timestamp': 17.25, 'end_timestamp': 45.26, 'index': 0}
@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},
}