Every African language deserves representation in AI

Open-source NLP models, datasets, and tools built for African languages. Try our demos right here, or dive into our research.

What we do

Building NLP for Africa

From speech recognition to phoneme conversion, we create tools that make it easy to work with African languages in NLP.

Language Identification

A CPU-friendly language ID model covering 1,386 African languages. It runs on a phone or laptop and requires just 5 seconds of audio.

Corpus Builder

Tools to access monolingual and parallel data for 693 African languages from multiple sources.

G2P Conversion

Rule-based grapheme-to-phoneme conversion for 400+ African languages, converting text to native orthography or IPA.

8 Open Datasets

Curated speech and text datasets freely available on HuggingFace, covering hundreds of African languages.

Live demo

Try It Now

Identify which African language is being spoken in an audio clip. Uses a fast version of Omnilingual ASR to transcribe speech, then a lightweight classifier names the language. Runs entirely on CPU.

African Speech ID

1,386 African languages

CPU-friendly: 3.5x realtime on one core
Classification head: 265 MB
Speech recogniser: 365 MB
5 seconds minimum audio
Audio ClassificationONNXApache 2.0
Live Demo

African Speech ID

Pick a sample audio below to see what the model makes of it.

Five seconds minimum, ten or more works best.

Open data

Open Datasets

Curated speech and text datasets covering hundreds of African languages, freely available on HuggingFace.

Africa Corpus

16.6M rows · 276 languages · Monolingual and parallel data

View Dataset →

African Speech IPA

776K rows · 934 languages · IPA phoneme transcriptions

View Dataset →

Open Bible Speech African

553K rows · 1.09K likes · Bible audio across African languages

View Dataset →

African Transcribed Speech

552K rows · 113 languages · Parallel and monolingual

View Dataset →

Open source

Our Projects

Tools built by the community, for the community. All code is open-source on GitHub.

africa-corpus-builder

14 stars

Get access to monolingual and parallel data for 693 African languages

View Repo →

africa-g2p

14 stars

Rule-based grapheme-to-phoneme for 400+ African languages

View Repo →

afrispeech-selector

14 stars

Easy access to speech data across 142 African languages for TTS and ASR

View Repo →

Get Involved

We welcome contributions from developers, linguists, and researchers working on African language technology.