Cuda Not supported for loading voices using torch.

[rkeshwani]

If trying to manually load a model to a cuda device and you load up a voice via torch.load to the device, pipeline.py doesn't like that the tensor type is a torch.cuda.TorchTensor. Specifically the load_voice method.

kokoro/kokoro/pipeline.py

Line 167 in dfb907a

def load_voice(self, voice: Union[str, torch.FloatTensor], delimiter: str = ",") -> torch.FloatTensor:

[rkeshwani]

You could switch to using is_tensor and torch.tensor as the data type agnosting of cpu or gpu?
https://alband.github.io/doc_view/tensors.html

I also don't know if this is the only place you need to make this change.

I don't know much about cpu or gpu specific development otherwise I would open my own pull request and do it myself.