Train a new Speaker

#1
by Chrisdercutter - opened

Thanks for your great model.

I tryed to finetune the pretrained German model with a new Speakerr, but i didnt manage to get anywhere. Maybe you could point me in a direction?

https://huggingface.co/blog/Steveeeeeeen/llasagna

The training code and a blog article is published @Steveeeeeeen . That may be your best bet. For a single speaker you can just finetune as is. For different ones you have to prepend a name of the speaker.

It could be benefitial to finetune from the german or other base model: https://huggingface.co/canopylabs/3b-de-pretrain-research_release

Thank you very much !

SebastianBodza changed discussion status to closed

Sign up or log in to comment