sdatkinson
[BREAKING] Don't require pre-silence in data
https://github.com/sdatkinson/neural-amp-modeler/blob/50df283d619f6940450a4eb00dcfb063c617b7b6/nam/data.py#L244C9-L244C34
I don't buy my argument because this is what the receptive field (nx) is for--the first output that we care about is far enough into the output audio that (we assume) everything required to predict it is present in the input.
Remove this.
This will break any code that uses that argument.
Ah, I remember:
Models have the ability to pad_start when making predictions. This prepends enough silence that the first sample of the output can be predicted for.
This is used during validation so that models with different receptive fields can compete "apples to apples" and not do better/worse because of how many of the first output samples they'd otherwise skip.
Given this (it's a validation thing), I don't think it's right to just remove it ("It must have been there for a reason"--and sure enough, now I remember the reason!)
But perhaps there's a better way to handle this.
Closing for now. I'll open something new (or hopefully find this w/ the search function) if it pops up again.