KoljaB/realtimetts ? reverse-engineered prompt

Build me a Python library that turns text into speech as it arrives, so it can start talking before the full text is finished. I want it to work with plain strings, generators, and streamed LLM output, and let me play the audio locally right away or save it to a WAV file instead. It should support a few different speech options, like a basic local system voice, some cloud based voices, and at least one local voice cloning option, with a simple fallback if one engine fails.

Please include a small example script, a clean way to switch voices, and controls for pause, resume, stop, and muted output. I also want it to be easy to use from a FastAPI app or another simple demo app. If you need to, look up the current docs online for any tricky engine setup details, but keep the final result easy for a normal developer to try quickly.