Steven Atkinson
Steven Atkinson
Write template
Possible upstream in NeuralAmpModelerCore
See #164. Catch2 seems reasonable. Opening this Issue to invite discussion. cc @Khalian
Same as https://github.com/sdatkinson/neural-amp-modeler/issues/277, though I'm not sure if Sphinx works for C++ (never tried).
Since an impulse response can be converted losslessly to a linear NAM, it'd be neat to be able to instantiate a Linear model using a .wav file.
# TODO checklist - [ ] Factory to initialize from `.nam` file. - [ ] Document `.nam` file version 0.5.5 to include `"sequential"`. - [ ] No `"weights"` nor `"sample_rate"`...
It would be nice for models to be defined as having an arbitrary number of channels as inputs/outputs. One thing that this would allow is for "(directed, acyclic) graph" NAMs...
Just a "little" task, but it'd be swell to take some of the TODOs around e.g. ring buffer handling and make good on them to rework the Conv1D class (cf...
This assertion might be incorrect: https://github.com/sdatkinson/NeuralAmpModelerCore/blob/846968710a670d662b15e449edba852d747d748e/NAM/wavenet.cpp#L159-L166 I don't think that it should be the receptive field of the layer array, but the r.f. of the single-largest layer. Worth thinking about...
Tanh uses a lot of compute (hence "fast tanh"). But this e.g.: https://github.com/sdatkinson/NeuralAmpModelerCore/blob/846968710a670d662b15e449edba852d747d748e/NAM/activations.h#L75-L81 should be able to be implemented more idiomatically with Eigen. It would also be nice for any...