Closed JackTemaki closed 1 year ago
Is the rationale to be as close as possible to the old implementation or just to replicate the behavior as good as possible?
Is the rationale to be as close as possible to the old implementation or just to replicate the behavior as good as possible?
In my opinion it could also be only replicating, so yes I can go over the code and use better names.
Thanks to @mmz33 , here some specaugment compatible to old Returnn/TF setups. Code follows mostly the Tensorflow implementation.