michsethowusu/kasanoma

Offline-first TTS models for African languages # Kasanoma – Offline TTS Models for languages of Africa **Kasanoma** is an open-source initiative dedicated to building and sharing **offline** **Text-to-Speech (TTS)** models for languages of Africa. With Kasanoma models, you can generate natural-sounding voices entirely on your local device — **no internet connection required**. All models in this project are built using Piper, a fast, lightweight neural TTS system that runs efficiently on laptops, desktops, Raspberry Pi, and other low-resource devices. You can also try out the models live online at our demo site. The models are trained for a limited amount of time due to GPU resource availability but they can be finetuned for better results using the training checkpoints which are also published. --- ## Available Models We currently provide offline TTS models for the following languages: | Language | Main Countries | Native Speakers | Total Speakers | |----------------|-------------------------------------------------------------|-----------------|----------------| | Twi (Akan) | Ghana | ~9–10 million | ~17 million | | Chichewa | Malawi, Zambia, Mozambique, Zimbabwe | ~12 million | ~18 million | | Makhuwa | Mozambique | ~7–8 million | ~8 million | You can download the models from releases. > More African languages will be added soon. > **Languages in the pipeline:** Kikuyu, Tshiluba, Amharic, Vai. --- ## How to Use the Models Offline All Kasanoma models are compatible with Piper. Here’s how to run them without internet access: 1. **Install Piper** You can install it via pip: ```bash pip install piper-tts ``` Or download a standalone binary from the Piper releases page for your platform (Linux, macOS, Windows). 2. **Download a Kasanoma Model** Get the `.onnx` model file and its matching `.json` config file for your language. …