Supertone
Supertone

Supertonic 3

Release Date 2026-04-29Open Model(OpenRAIL-M)

Supertonic 3 is a lightweight open-weight speech-synthesis model from Supertone. It runs on ONNX Runtime entirely on your own device, so synthesis needs no cloud call. It is designed to run quickly on a CPU without a GPU, which makes browser and edge deployment easier, and the company describes it as far smaller than the larger open TTS systems in common use. Compared with the previous release it expands support to 31 languages, including Korean and Japanese, cuts repeat and skip failures on both short and long sentences, and improves speaker similarity. Simple expression tags such as laugh, breath and sigh are supported. The open package ships only fixed preset voices; a custom voice requires a style file made with Supertone's paid Voice Builder. For a hosted option without running the model yourself, Supertonic 3 is also offered in Supertone Play and the Supertone API. The weights are released under the OpenRAIL-M license.

API cost