CorentinJ/Real-Time-Voice-Cloning ? reverse-engineered prompt

Build me a local Python voice cloning tool like the SV2TTS demo. I want to give it a short audio clip, around five seconds if possible, have it learn the speaker voice, then type any sentence and generate speech that sounds like that speaker. Please include a simple desktop toolbox where I can load or record audio, preview the voice, enter text, synthesize it, and save the output audio. Also include a command line demo for the same flow, since I may run it without the interface.

Set up the project so it can run on Windows or Linux, with an easy install using uv, ffmpeg for audio loading, and support for either NVIDIA GPU or CPU. Use pretrained models by default and download them automatically if they are missing. Keep the code organized for encoder, synthesizer, and vocoder parts, and add basic tests or sanity checks so I know the setup worked. Look up current docs online if you need to.