Datasets:
dataset_info:
features:
- name: audio
dtype: audio
- name: transcription
dtype: string
splits:
- name: train
num_bytes: 109304535928.432
num_examples: 90244
- name: validation
num_bytes: 1051506219.236
num_examples: 1013
- name: test
num_bytes: 1226193261.48
num_examples: 1020
download_size: 93176985982
dataset_size: 111582235409.148
license: cc-by-nc-4.0
task_categories:
- automatic-speech-recognition
- audio-classification
tags:
- automatic-speech-recognition
- audio-classification
- Portuguese
- ASR
language:
- pt
pretty_name: mTEDx PTBR
Multilingual TEDx (Portuguese speech and transcripts)
NOTE: This dataset contains only the Portuguese portion of the mTEDx dataset, already processed and segmented into parts.
Multilingual TEDx (mTEDx) is a multilingual speech recognition and translation corpus to facilitate the training of ASR and SLT models in additional languages.
The corpus comprises audio recordings and transcripts from TEDx Talks in 8 languages (Spanish, French, Portuguese, Italian, Russian, Greek, Arabic, German) with translations into up to 5 languages (English, Spanish, French, Portguese, Italian). The audio recordings are automatically aligned at the sentence level with their manual transcriptions and translations. Each .tgz file contains two directories: data and docs. docs contains a README detailing the files provided in data and their structure. Test sets for all IWSLT 2021 language pairs can be found in mtedx_iwslt2021.tgz. For more information on the dataset please see the dataset paper.
Contact: Elizabeth Salesky, Matthew Wiesner. esalesky@jhu.edu, wiesner@jhu.edu
Citation: If you use the Multilingual TEDx corpus in your work, please cite the dataset paper:
@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},
}