Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Voice but not checkpoint for italian? #627

Open
kirys opened this issue Oct 9, 2024 · 12 comments
Open

Voice but not checkpoint for italian? #627

kirys opened this issue Oct 9, 2024 · 12 comments

Comments

@kirys
Copy link

kirys commented Oct 9, 2024

Hi all
I'm using pipertts for a small experiment at home, I'm Italian so I'm using the Italian voice "Paola".
I was thinking of training another voice, so I started searching tutorials, but the ones I found starts from a checkpoint.
But in the repository i don't see an Italian checkpoint
https://huggingface.co/datasets/rhasspy/piper-checkpoints/tree/main
What should I do?
Thank You
K.

@Milkom121
Copy link

same problem here

@GianlucaApollaro
Copy link

Hi. I have the same problem. I would like to train an Italian voice, but because I don't have a check point everything sound like gibberish.

@kirys
Copy link
Author

kirys commented Oct 29, 2024

I tried using the Spanish checkpoint, but the results were weird (and wasted 10€ of colab credit ^_^).
I wonder how much GPU time would take to make a Italian checkpoint

@GianlucaApollaro
Copy link

Hi. I could finally make an Italian check point based on an English one and it worked quite well. Sometimes it gets weird, but I think that I could use a longer dataset to train even better. I'm not using colab, but I have my own GPU.

@kirys
Copy link
Author

kirys commented Nov 5, 2024

Mind sharing some info like
How many minutes/hours of voice did it take?

@GianlucaApollaro
Copy link

Hi, for the moment I used around 40 minutes, and it doesn't sound that bad. I am planning to get to 2 hours of material in the near future.

@TeoColuccio
Copy link

TeoColuccio commented Nov 11, 2024

Hi. I could finally make an Italian check point based on an English one and it worked quite well. Sometimes it gets weird, but I think that I could use a longer dataset to train even better. I'm not using colab, but I have my own GPU.

Hi, did you used the Lessac checkpoints and fine-tuning with italian voices?

@GianlucaApollaro
Copy link

yes, it worked quite well, once I found the right amount of material.

@aventrax
Copy link

Hi guys, I would like to have another Italian Voice too, could you share your results? The project has only 2 voices but with strong accent that I don't like much.

@kirys
Copy link
Author

kirys commented Dec 17, 2024

I'm working on it, I'm about at 1700 epoch I'll publish on hugging face as soon as it reach 2000

@GianlucaApollaro
Copy link

Hello, I'm using a voice of a friend who's a voice actor, but unfortunately I don't have permission to release it. I'll try to find some free voices that I can use.

@kirys
Copy link
Author

kirys commented Dec 18, 2024

I've almost finished training the first voice based on the dataset I published here
https://huggingface.co/datasets/kirys79/piper_italiano
It has nice quality but an accent of south italy (Siciliano), but should be much better for fine tuning than any other language.
Tomorrow I'll load it on this repo
https://huggingface.co/kirys79/piper_italiano

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants