Data Infrastructure
Crowdsourcing Language Data for Inclusive AI
Native speakers building the language data AI is missing
All Voices lets native and fluent speakers translate, transcribe, record, describe and review language data, building trusted datasets for languages that technology still leaves out.
The challenge
More than 7,000 languages are spoken around the world, yet most have very little data behind them, so AI cannot understand them well. African languages make up less than 1% of the data used to build modern AI. Without high-quality, trusted data, speech recognition, translation and assistants simply do not work for most people on the continent.
Our solution
All Voices puts the people who speak these languages at the centre of building the data. Contributors take on five kinds of tasks:
Translate source text into their language.
Transcribe audio into text.
Record speech, with automatic audio quality checks.
Describe images by speaking or writing.
Validate each other's work through peer review.
Contributors earn points and streak bonuses, compete on leaderboards and help preserve their linguistic heritage. Partner organisations can upload their own source material, choose the task types and receive a reviewed dataset, exportable as JSON or CSV.
Impact
All Voices is part of the data foundation behind Mansa, which is built on 100B+ curated tokens and 13,000+ hours of expert-validated speech. Every validated contribution makes language technology work better for the people who speak these languages.
Get involved
Contribute through the All Voices app on iOS and Android or on the web, or talk to us about running a data project for your organisation.
