xiph/rnnoise — reverse-engineered prompt

Build me a small C project for speech noise reduction using an RNN, basically the RNNoise style setup. I want it to compile from source, pull the model files automatically if needed, and give me a simple command line program that takes a noisy raw 16 bit mono PCM file at 48 kHz and writes out a cleaned raw PCM file. Please make it usable as a library too, not just a demo app, and keep the example easy to run.

If the repo already includes training scripts, keep those working as well so someone can generate features, train a model, export the weights back to C, and rebuild with the new model. If there is support for loading model files at run time, include that too. Add a short README with exact build and run steps, plus any important input format gotchas since this is raw PCM and not WAV. Look up current docs online if you need to.