svc-develop-team/so-vits-svc ? reverse-engineered prompt

Build me an offline singing voice conversion app that can take a vocal track, let me train or load a singer model, and turn the source singing into that voice while keeping the melody and pitch feel intact.

I want a simple web interface where I can upload audio, choose a model, run conversion, and download the result. It should also support the full workflow for preparing data, preprocessing it, training a model, and then doing inference later. If it helps, add a way to export for easier deployment, and make the whole thing work well for local use on a normal machine.

Please make the experience as smooth as possible, with clear buttons, progress feedback, and sensible defaults. If you need current setup details or model compatibility info, look up the latest docs online before wiring things up.