Help with Brazilian Portuguese

[edivanteixeira]

I would like to train voices but in another language, I looked in the documentation but it seems to me that the project only includes English, can anyone help me?

[neonbjb]

Happy to help anyone train this in a different language as long as they commit to open sourcing it. Step 1 would be gathering a large dataset of speech, at least 10,000 hours. If your language has a wav2vec2 speech recognition model available (PR does), then the speech does not need to be transcribed.

Once you have this, reach out and I'll try and help you do further training

[cobayashi7411]

It's a loose figure. Tortoise was trained on ~50k hours of speech. Based on the model size and the number of epochs to convergence (2), I believe it underfit.

I am sure you could train a model like this on less than 10k hours but you'll need to be very cognizant of overfitting. More importantly, if you have less than 10k hours, I would worry about the number of speakers you have in your dataset.

I would like to try it. I would like to train it in my language. PLS.

[Erim32]

Hello @neonbjb, is it possible to update the requirement for 2023 version ?
Just to have an idea, how many Go dataset represents for one language ?

Thk you

[fabioravila]

In our language, it is really difficult to find quality datasets. So creating a giant dataset is a very complicated task.
But if you get as much automation as you can, automating the entire process would make life a lot easier.

I'm working on a drag-n-drop style pipeline where we would simply place the audio or video sources and be able to separate the good from the bad and filter a good final collection.

I will send you an email requesting these weights to help with my pipeline.

[RahulBhalley]

Where is the training script???

[felipeduartedeveloper]

I'm also looking for support for Brazilian Portuguese

[dreamfaceprod]

I would like to train voices but in another language, I looked in the documentation but it seems to me that the project only includes English, can anyone help me?

Hello, Edivan!
do you have success? if yes, could we cooperate?

[josplit]

I would love if this worked in Spanish (in my test the subject actually does speak Spanish but with a weird French accent haha but it actually worked) is there any way we can help you with other languages? your work is incredible, thank you.

[djzigoh]

hi!! any updates on this? thanks!

[douglasescrito]

Eu também gostaria de ver essa ferramenta em portugues brasileiro.

[tiwale01]

Afternoon, any updates on this post? Was anything done ??

[alexandreespejo]

Any updates on this ?

[carlosedubarreto]

I'm not sure if it worth pursuing this work of making a brazilian one. I found another repository that already has some voices in pt-br that re great. its bark https://github.com/suno-ai/bark

[alexandreespejo]

Thanks a lot @carlosedubarreto, do you know if with bark is possible to train new voices models ?

[carlosedubarreto]

@alexandreespejo , the bark repo itself doesnt have, but the author suggest another one that you could use.

more info on this issue
suno-ai/bark#37

[jjjota23]

achei o suno muito robótico... acho que vale a pena treinar o modelo acima