k2-fsa/OmniVoice ? reverse-engineered prompt

I'd like to build a powerful text-to-speech application in Python. The main thing it should do is high-quality voice cloning. I want to be able to upload a short audio file of a person speaking, and then have the tool generate new speech in that same voice from any text I provide.

It should also have a "voice design" feature where I can just describe the voice I want using attributes like 'female, low pitch, British accent' and it will create it without needing a reference audio. A key requirement is that it must support a huge number of languages, ideally over 600. It would also be cool if I could add non-verbal sounds like [laughter] into the text.

Please package this up so I can run a simple web demo on my local machine to try everything out. Use a modern, fast model for this, maybe something from Hugging Face if that works.