SayaSS / vits-finetuning

Fine-Tuning your VITS model using a pre-trained model
MIT License
551 stars 86 forks source link

How was the n-speaker model created? #43

Open ikpark09 opened 1 year ago

ikpark09 commented 1 year ago

Thank you for providing a good helpful repo.

I have a single-speaker model that has been trained for a long time. However, when I try to train a multi-speaker model using it, I encounter errors.

The model you provided has both single-speaker and multi-speaker. How was this achieved?