Skip to content

Dataset · legacy-datasets

Multilingual Common Voice speech dataset

legacy-datasets/common_voice

7,335 hours of verified speech in 60 languages from Mozilla for ASR training.

What's inside

Common Voice is a Mozilla initiative that helps teach machines to understand spoken language. The dataset currently contains 7,335 hours of verified speech in 60 languages, with the list of languages continuing to grow. Metadata lists languages such as ab, ar, as, br, ca, cnh, and others. The task is automatic-speech-recognition.

What you can do with it

  • Training ASR models
  • Fine-tuning multilingual speech models
  • Evaluating ASR on diverse and low-resource languages
Tasks
Automatic speech recognition
Languages
Abkhazian, Arabic, Assamese, Breton, Catalan, cnh, Czech, Chuvash and 4 more
Size
100K–1M rows
Downloads this month
1.2K
Likes
148

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("legacy-datasets/common_voice")