Appropriate model size

[34j]

I know the model parameters have not changed much from the original VITS (for 100 speakers), but is it appropriate to keep the model size to train a model for one speaker?

[Lordmau5]

I did actually want to ask about the model parameters as I'm personally completely unsure what could be changed to alter training in both a good and bad way.

learning_rate for example. Is that good to have on the default 0.0001 or would 0.0005 be better, perhaps even 0.05?

As for all the other stuff further down in the "model" section, such as inter_channels, hidden_channels, etc., I have absolutely no idea.

I couldn't find any documentation about those parameters :/

[34j]

I think learning_rate is currenly too high for a small dataset, but it might also be related to batch size