Help every African language be heard.

Write code

Kuma is a desktop app built with Rust and React on Tauri, using whisper.cpp for transcription. To run it locally you need Rust, Node, cmake and ffmpeg:

git clone https://github.com/geniuskidkanyi/kuma
cd kuma
npm install
npm run tauri dev

The first build compiles whisper.cpp from source and takes a few minutes. After that, pick up an open issue, fix a bug, or open one to propose a feature. Small, focused pull requests are easiest to review.

Train and convert models

This is the highest-impact contribution. If you can fine-tune Whisper on a language, or convert an existing fine-tune to GGML, you unlock that language for everyone using Kuma. The models guide walks through importing, converting and fine-tuning step by step.

When you publish a model, tell us in an issue so it can be listed in the in-app African languages catalog.

Record and validate voice

Models are only as good as the speech they learn from, and most African languages have very little open recorded data. You do not need to be technical to help:

• Record and validate sentences in your language on Mozilla Common Voice.
• Join the African NLP community at Masakhane, which coordinates datasets and research across the continent.
• Export from Kuma: anything you transcribe and correct can be turned into a clean dataset with Contribute dataset, then donated.

Translate the app

A transcription tool for African languages should not have an English-only interface. If you would like to help translate Kuma's own screens into your language, open an issue to say which language you can cover, and we'll set up the strings for you to work on.

Spread the word

Tell journalists, newsrooms, universities, radio stations and NGOs who work in African languages that a free, offline option exists. Star the repo so more people find it.