Skip to content

Dataset · fsicoli

Mozilla Common Voice 17.0 speech corpus mirror

fsicoli/common_voice_17_0

Large multilingual audio dataset for speech recognition converted from Mozilla Common Voice version 17.

What's inside

An unofficial copy of the Mozilla Common Voice Corpus 17.0 dataset, downloaded and repackaged from the official project site. Total volume ranges from 100 billion to 1 trillion elements. The corpus covers dozens of languages, including Abkhaz, Albanian, Arabic, Armenian, Bashkir, Belarusian, Chuvash, and many others.

What you can do with it

  • Training and fine-tuning ASR models for multiple languages, including low-resource ones
  • Benchmarking speech recognition quality across diverse accents
  • Pre-training multilingual acoustic encoders
Tasks
Automatic speech recognition
Languages
Abkhazian, Afrikaans, Amharic, Arabic, Assamese, Asturian, Azerbaijani, Bashkir and 4 more
Size
100B–1T rows
Downloads this month
32.1K
Likes
19

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("fsicoli/common_voice_17_0")