Voice but not checkpoint for italian?

[kirys]

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]

same problem here

[GianlucaApollaro]

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]

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]

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]

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

[GianlucaApollaro]

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]

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]

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