I've reconstructed everything I'd wanted so far, but I'll keep an eye on this. This also has an option to learn from just a recording of its test sweep (like the technique I used). I suppose a learned model could be useful for nonlinear modes, allowing a more complex function when my current approach fails (which is quantifiable with the existing tests).