Training for Italian and German works, but Portuguese and Russian do not.

[frixos25]

I am trying to train a model in German, Italian, Portuguese, and Russian, but only Italian and German produced successful results. The training for Portuguese and Russian resulted in unusable model files. The phoneme extractor works fine in all these languages. What can I do to successfully train the model for Portuguese and Russian? I also modifed the files portuguese.py and portuguese_bert.py with the link model_id = 'neuralmind/bert-base-portuguese-cased' and the same in the other languages. The phonemes seems to be accurate.

[yukiarimo]

Will update you when implemented. I’m currently working on rewriting everything, so you can check out my YunaTTS repo

[gdurifw]

I am trying to train a model in German, Italian, Portuguese, and Russian, but only Italian and German produced successful results. The training for Portuguese and Russian resulted in unusable model files. The phoneme extractor works fine in all these languages. What can I do to successfully train the model for Portuguese and Russian? I also modifed the files portuguese.py and portuguese_bert.py with the link model_id = 'neuralmind/bert-base-portuguese-cased' and the same in the other languages. The phonemes seems to be accurate.

hi, can you explain how to do the training to add the Italian language? Is preprocess_text.py enough or do we need to create a new structure starting from es_phonemizer? Ciao :)

[ghost]

Hello, I am also interested in how to train the model on an Italian dataset. We have a dataset that includes a set of 1,000 audio files, and we have modified the repository's code to ensure it follows the guidelines of other languages. Are we heading in the right direction? @frixos25 Could you give us some advice on how you did it?

Thank you very much in advance.

[yukiarimo]

1. If language does not exist you need to just implement it (find phoneme logical somewhere)
2. Otherwise, we can just simply transliterate into English and train a new speaker (this works)!

[julia-imlauer]

@frixos25 could you share the code for the German training? I would be interested in training German model too.

[DiamondGlassDrill]

Hi everyone,

I’m also very interested in seeing a German speaker added. Even better, a step-by-step guide on how we could train our own German TTS model would be incredibly helpful for the community.

If anyone has pointers—whether it’s to data preparation, training scripts, or fine-tuning tips—please share them.

Thanks so much in advance!