Adding Turkish language support

[g-hano]

I created turkish.py and turkish_bert.py to support Turkish language. Used dbmdz/bert-base-turkish-cased as tokenizer

[kadirnar]

@g-hano Did you train with Turkish data?

[g-hano]

@g-hano Did you train with Turkish data?

Yes I did, had to update many lines but finally was able to train on 4 T4s. Extra training on 46k audio samples made the model produce clearer outputs.

[tosunozgun]

Hello, thank you for supporting the Turkish TTS model, first of all. When I reviewed your code, I noticed that you convert text to lowercase during normalization. Unfortunately, for Turkish, uppercase "I" is converted to lowercase "i," which is incorrect. As a solution, the following change is needed: text.replace("I", "ı").lower(). Since you are using lowercase text, I recommend trying ytu-ce-cosmos/turkish-base-bert-uncased.

[g-hano]

Hello, thank you for supporting the Turkish TTS model, first of all. When I reviewed your code, I noticed that you convert text to lowercase during normalization. Unfortunately, for Turkish, uppercase "I" is converted to lowercase "i," which is incorrect. As a solution, the following change is needed: text.replace("I", "ı").lower(). Since you are using lowercase text, I recommend trying ytu-ce-cosmos/turkish-base-bert-uncased.

Thank you for your recommendations, I updated the code.

[furkeen]

Thanks for the great contribution @g-hano . I'm trying to test it however I couldn't find a way to test. Is there any way to test something like that?
melo "Merhaba ben Türkçe konuşabiliyorum." output.wav --language TR

[kadirnar]

I am developing the MeloPlus library to add more features to the MeloTts library. I added Thai and Turkish language support.

@furkeen You should find a Turkish model trained with the melotts model.

[karayakar]

Any huggingface space for testing? or sample output?