Supertone Supertonic 3 expands on-device text-to-speech from 5 to 31 languages
Supertone released an ONNX-based text-to-speech model with about 99M parameters. It supports 31 languages and expressive laughter, breath, and sigh tags, and reduces repetition and omission errors compared with v2. The 404 MB of ONNX assets are available on Hugging Face and run locally on iOS, Android, and HarmonyOS.
Visit original source