adds wandb logging for all relevant training parameters
logs generated audio files to wandb after each step (could maybe be a lot, i just went with the defaults they implemented -> but we can reduce that in the future)
after each step logs unconditional generated audio with a fixed seed