Thanks. It's an interesting jumping off point to several projects on github with links to papers in the subject. The papers give more of a sense of the computational requirement both for training and for running the model as an effect once trained. I'd like to track down the code for the RNN model as it seems to be more efficient for a given level of performance than the WaveNet model used in that project, at least according to
this paper that's linked in the
PedalNetRT page. Even then it's significantly more computationally expensive than room correction or crossovers, at least when modeling highly distorting amplifiers and pedals to a high level of accuracy.