More than 1 speaker per language?

[jidkano]

Hello, would it be possible to add a default of more than just 1 speaker per language? Maybe 3-4 (2 males, 2 females) per language would be great.

[yukiarimo]

Yes. When training a dataset like this:

path|speaker1|lang|text

So, you can do:

path|speaker1|lang1|text path|speaker2|lang1|text path|speaker3|lang1|text

[yukiarimo]

Hey, just checked. Don’t do this! If you want same speak to speak different languages - create new speaker, but use same voice but with different language. It is the best approach!

[youngandbin]

@yukiarimo I also wanna add more speakers. You mean it is the best to do like this?

path|speaker1|lang1|text path|speaker2|lang2|text path|speaker3|lang3|text

[yukiarimo]

Exactly!

[luckycontrol]

@yukiarimo Hello, I'm doing finetuning to add korean speaker. I have a question. How long should I train for?

[yukiarimo]

Minimum data -> 1 hour

Good data -> 10 hours

Enough data -> 24 hours

How long? -> Generate audio each 10k steps. Stop the training whenever you like the output

[luckycontrol]

@yukiarimo Thank you very much for your reply