Model fine tuning

[lbdave94]

Has anyone been able to finetune the model on their dataset? I tried on proprietary dataset (very good quality audio at 44.1-48kHz, expressive) but could not improve the final result; indeed some artefacts are introduced [I run stage 1 loading enhancer from pre-trained release model and then stage 2].

I was wondering if anyone managed to fine-tune the model on their own dataset. Because I'm getting the doubt that it can't be done without having the discriminator weights.