NeuTTS is a family of open-source, on-device text-to-speech models from Neuphonic designed to run anywhere — phones, laptops, even a Raspberry Pi. The project has about 6,300 stars on GitHub and was updated in 2026.
Most modern TTS lives in the cloud; NeuTTS goes the other way. Its speech language models are small enough for edge hardware: NeuTTS Air has roughly 360 million parameters, the multilingual NeuTTS Nano about 120 million, and NeuTTS 2E adds emotional expression. All three support instant voice cloning from as little as three seconds of reference audio, and GGUF quantization lets the models run efficiently on phones and single-board computers. The family is built on NeuCodec, a 50Hz neural audio codec, and generated speech carries a watermark for provenance. One caution from the project itself: neutts.com is not the official site — the models live on GitHub and neuphonic.com. The models are distributed through Hugging Face with example code for several runtimes, which makes it straightforward to prototype on a laptop and then move the same model to a phone or an embedded board. For accessibility tools and offline assistants, that local, private operation is the main attraction.
The NeuTTS models are free and open to use under their respective licenses. There is nothing to pay to download and run them locally.
Developers adding offline voice to apps, wearables, kiosks, and embedded products — anywhere a cloud TTS round-trip is too slow, too costly, or a privacy problem.