rikorose/deepfilternet ? reverse-engineered prompt

Build me a speech noise suppression tool that can clean up noisy 48 kHz wav files using a pretrained model.

I want a simple command line app where I can drop in one or more wav files, choose an output folder, and get back enhanced audio. It should support a post filter option and a way to compensate for processing delay. If it helps, make it possible to run with the default model or point to a different model folder.

Also add a small demo screen so I can try it interactively on Linux, and make sure there is a clean Python way to use it from my own scripts. If there are any current docs or model details you need, look them up online and follow the existing project approach.

Keep it focused on real time speech enhancement, and make sure the whole thing works end to end with the pretrained weights included in the repo.