Finetuning and Vocoder Training

[sulaiman-kagumire]

Thank you for this great work!

I successfully trained on my custom dataset for another language. In the saved checkpoints folder, I see all checkpoints, including the last.ckpt. To sysnthesize speech using my custom trained model, do I need to train the vocoder? because I have another language? or i can use the already pretrained one with matcga-tts?

[shivammehta25]

You should be able to use the universal version of the vocoder. But if it is too noisy, you can try switching to BigVGan or fine-tuning the vocoder entirely.

[sulaiman-kagumire]

Oh okay. I will try out the universal version. Thank you