any advantages to using a custom DDSP model for processes 4.0 / 5.0?
Opened this issue · 1 comments
markhanslip commented
Thanks for the great work, is super impressive and useful.
I was wondering if there would be a likely improvement in results if the DDSP model is trained on the same data as the diffusion vocoder in versions 4.0 / 5.0? Or would it be a waste of effort?
yxlllc commented
Do you mean making a pre-trained model? It is of little significance for the pure ddsp model. Unlike diffusion models, DDSP models have relatively limited capabilities and cannot be scaled up effectively.