I'm new to the field. I want to ask, is the training set the "raw" version. Generally speaking, deepfake Settings, such as SRM, SPSL, etc., are trained in the "raw" version, and then tested in the "c23" and "c40" versions???