You can reduce the size of mini-batch, however, it would take a lot of time to train VoxCeleb2 using a single GPU
(regardless of your GPU generation, normally many recent systems take days using 4 GPUs )
My quick advice is to train the model using the train data of VoxCeleb1 and evaluate using VoxCeleb1 testset.
First of all thanks for such amazing work of RawNet and its variants!
I just want to know about the hardware requirements for training RawNet2 and modified RawNet2. I have following two questions: