I was going through the MPD implementation in your DAC repository (discriminator.py#L46) and noticed that pad_to_period uses nn.functional.pad with mode="reflect".
Was reflect padding chosen empirically because it works better in practice, or is there a specific signal processing reason behind it?
I was going through the MPD implementation in your DAC repository (discriminator.py#L46) and noticed that pad_to_period uses nn.functional.pad with mode="reflect".
Was reflect padding chosen empirically because it works better in practice, or is there a specific signal processing reason behind it?