Every voice, every dialect, forever.

AI only speaks the languages it was trained on — and African languages are barely in the training data. Afrolingua Hub exists to change that by building the data itself, contribution by contribution:

• Old books and written text become language-modelling corpora — what large language models read to learn a language.
• Voice recordings with transcripts train automatic speech recognition (ASR) and text-to-speech (TTS) — machines that hear and speak your dialect.
• Dialect–English sentence pairs train machine translation (MT), the bridge between your language and the world.

Because the dataset is open and dialect-labelled, any researcher, startup or community group can build with it — and the credit always flows back to the speakers who gave it.

Every contribution is licensed Creative Commons Attribution 4.0 (CC BY 4.0) at the moment it’s accepted. We store the exact consent text you agreed to, and when, alongside your contribution. That license is irrevocable by design: no company — including us — can ever enclose the dataset, paywall it, or sell exclusive access to it.

Attribution means your name (or chosen display name) travels with your words. Researchers who build with the corpus credit the people who gave it.

A contribution here is never just “Yoruba” or “Igbo.” It is Ìjẹ̀bú Yoruba or Standard Yoruba; Onitsha, Owerri, Nsukka or Enuani Igbo; Sokoto or Kano Hausa. The dialect label is what makes the dataset genuinely useful — and what makes every community’s effort visible on its own terms.

Big languages and small ones are equally at home. Even the largest languages hold under-recorded dialects, unscanned hymn books and oral registers no model has ever heard. And if your language or variety isn’t on our list, you’ll be able to propose it — the map should never wait on us.

Approved contributions earn coins, badges and leaderboard standing. Today, coins are recognition, not currency. They are a public, auditable record of what you’ve given your language — every coin event is stored in an append-only ledger.

If and when funding exists for a language — a sponsor, a grant, a donation pool — we may open a redemption window that converts coins to cash rewards for that language, at a rate set at that time from that pool. Until then we will never promise a fixed cash value, and coins are not wages or an investment. When that changes for your language, you’ll see it clearly labelled here first.

Donations run through Flutterwave, with every payment verified server-side. The public transparency ledger shows the totals straight from the database, and every language fund is tracked on the support page. Sponsoring a language should feel as accountable as it feels good.