Speech to text for every African language. Offline, free, and open.
Kuma transcribes audio and video on your own computer using open Whisper models. Import a community-trained model for your language, fix the transcript, and export it. Nothing leaves your device.
Kuma means "word" or "speech" in Mandinka.
The problem
Most speech-to-text tools don't understand most African languages.
Stock Whisper supports only a handful of African languages, and it handles them poorly. Dozens of major languages, spoken by millions of people, aren't covered at all.
The reason is simple. The models were trained on very little recorded speech in these languages. A journalist in Banjul or a researcher in Kampala ends up typing interviews by hand.
That changes when the people who speak these languages can build and improve the models themselves.
How Kuma helps
A transcription app that grows with its community.
Offline & private
Runs entirely on your device. No internet, no API keys, no per-minute fees.
Bring any model
Import community fine-tunes for your language from a file or a Hugging Face link. See the guide →
Edit & export
Follow the audio with a highlighted transcript, fix it line by line, then export SRT, VTT or TXT.
Give back
Turn your corrections into an open dataset that makes African speech recognition better for everyone.
Who it's for
Journalists, researchers, students, radio and media houses, NGOs, podcasters, courts, and anyone documenting interviews or oral history in an African language.
Join us
Help every African language be heard by machines.
Kuma is built in The Gambia and maintained in the open. There is a place for you whether you write code, train models, speak a language, or translate. See all the ways to help →
Write code
Help build Kuma in Rust and React. Pick up an open issue or propose a feature.
Browse the issues → ModelsTrain models
Fine-tune Whisper on your language and convert it to a model anyone can import.
Read the guide → VoiceValidate & record
Native speakers: check output and contribute your voice as open data.
Start on Common Voice → LanguageTranslate the app
Help Kuma speak your language in its own interface.
How to help →Questions
Is it free?+
Yes. Kuma is free and open source under the MIT license. There are no accounts, subscriptions or per-minute fees.
Does it need internet?+
No. Transcription runs entirely on your computer. You only need a connection to download the app or a new model.
Which languages does it support?+
Any language that has a compatible Whisper model, including community fine-tunes. The list grows as contributors train and share new models. See how to add one.
Can I use it for my radio station or NGO?+
Yes. Kuma is free for personal, organisational and commercial use under its open-source license, and nothing you transcribe is sent anywhere.
How do I add my language?+
Import an existing community model by file or Hugging Face link, or fine-tune and convert one. The models guide walks through both.