loudkit

loudkit is an open-source text-to-speech engine that runs on your own hardware. It ships two models, loudr-1 and the faster loudr-1-turbo, with 28 voices across ten languages (Danish, Dutch, English, French, German, Italian, Polish, Portuguese, Spanish, Swedish), and it can clone a voice from about ten seconds of audio into a small profile file that works with both models.

It is built for developers and self-hosters: a Python library and CLI, native SDKs for Swift, Go, Rust and TypeScript that run without Python, PyTorch, ONNX Runtime and CoreML backends (CPU, NVIDIA GPUs, Apple silicon), and a local server with an OpenAI-compatible speech endpoint, so apps and agents that already talk to OpenAI's speech API can point at loudkit instead. After the model download, nothing leaves the machine.

Code and model weights are Apache-2.0.