Mozilla Common Voice
Mozilla Foundation
Volunteers reading sentences aloud in 100+ languages, donated to the public domain.
Crowdsourced recordings of volunteers reading short sentences, validated by other volunteers' votes, with optional self-reported age, gender, and accent. It is the largest public-domain multilingual speech set and the main resource for speech recognition in many smaller languages. Since October 2025 downloads come through the Mozilla Data Collective platform.
Access
- Cost
- FreeFree, but downloads require a Mozilla Data Collective account and acceptance of its terms.
- License
- Public domainCC0 1.0
- Sign-up
- Free account
- How you get it
- Bulk downloadMP3, TSV
What's in it
- One row is
- One recorded sentence clip with its text, up/down votes, and optional speaker demographics
- Covers
- Volunteers worldwide, in 100+ languages
- Years
- 2017-present
- Updated
- Quarterly
- Size
- Tens of thousands of hours across all languages; varies by release
Quality
Open, multilingual, and community-validated; recording quality and speaker balance vary widely by language.
- A few prolific speakers dominate some languages
- Demographics are optional and self-reported
- Read sentences, not natural speech (a separate Spontaneous Speech set began in 2025)
Connects to
- LibriSpeech ASR Corpus: English read-speech benchmark to compare accents and recording conditions
- OPUS Parallel Corpora: Text resources for the same lower-resource languages
Questions it could answer
- Which languages have the most validated hours per speaker?
- How balanced by gender and age is each language's data?