TungX/tts-supertonic ? reverse-engineered prompt

Build me a fast, local text to speech app and SDK that runs entirely on the user’s device and can work offline.

I want it to take text, let me choose a voice style, and generate clear 44.1 kHz audio in multiple languages, with an option to auto detect the language when I’m not sure. It should feel lightweight and quick to start, with no cloud calls and no privacy concerns. Please include simple examples for Python first, plus working sample apps for web, Node, and a few other common platforms if that fits. I’d also like support for a few expression tags like laugh, breath, and sigh so the speech sounds more natural.

Set it up so the model files can be downloaded automatically the first time, and make sure there’s an easy way to save the generated audio as a WAV file. If you need to check current ONNX Runtime or platform docs online, go ahead and do that.