Skip to content

Dataset · facebook

Multilingual corpus for speech translation and ASR

facebook/covost2

A large audio corpus based on Mozilla Common Voice for speech recognition and translation between English and dozens of languages.

What's inside

The dataset, ranging in size from 100 thousand to 1 million records, includes spoken translations from 21 languages into English and from English into 15 languages (metadata notes Arabic, Catalan, Welsh, German, Spanish, Estonian, etc.). Audio files are collected from the Mozilla Common Voice crowdsourcing project and stored in .mp3 format.

What you can do with it

  • Training end-to-end speech translation systems (Speech-to-Text Translation)
  • Training and evaluating multilingual ASR models
Tasks
Automatic speech recognition
Languages
Arabic, Catalan, Welsh, German, Spanish, Estonian, Persian, French and 4 more
Size
100K–1M rows
Downloads this month
600
Likes
52

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/covost2")