UniPC has been impressive in generating diffusion audio for both singing and speaking. However, the current UniPC implementation in the diffusers repo only supports 2D dimensions. Adding 1D support to the UniPC scheduler would be a great enhancement.
UniPC has been impressive in generating diffusion audio for both singing and speaking. However, the current UniPC implementation in the diffusers repo only supports 2D dimensions. Adding 1D support to the UniPC scheduler would be a great enhancement.
Example projects for reference: