Add XTTS Fine tuning gradio demo

[Edresson]

This PR has the propose of making XTTS GPT encoder fine-tuning more accessible to beginners users.

It implements a gradio demo that do:

1. Preprocessing of the uploaded audio or audio files in 🐸 TTS coqui formatter
2. Train the XTTS GPT encoder with the processed data
3. Inference support using the fine-tuned model

[Edresson]

@WeberJulian Did a Colab notebook to make it accessible for users that do not have GPU: https://colab.research.google.com/drive/1GiI4_X724M8q2W-zZ-jXo7cWTV7RfaH-?usp=sharing

We are also planning to do a video tutorial to make it easier for the beginner's users. As soon it is done we can add it on the docs and then merge it :).

[FurkanGozukara]

i was planning a tutorial and even had prepared auto data preparer and such haha :d

[WeberJulian]

i was planning a tutorial and even had prepared auto data preparer and such haha :d

it's still good to have multiple ways of doing the same things, especially since it might not have the same features.
I see you have an install_TTS.bat, that would be a great addition to the main repo too.

[FurkanGozukara]

awesome merged into dev. i was waiting this

[daswer123]

Guys, made a modification webui that was comfortable to use locally
https://github.com/daswer123/xtts-finetune-webui

[FurkanGozukara]

Guys, made a modification webui that was comfortable to use locally https://github.com/daswer123/xtts-finetune-webui

i think they should merge your pull request all positive changes

[lpscr]

hi @Edresson
thank you for the you work,
this working with other languages it's not in list like Greece ? just to know if possible right now or need wait updates ,

thank you

[Edresson]

hi @Edresson thank you for the you work, this working with other languages it's not in list like Greece ? just to know if possible right now or need wait updates ,

thank you

Hi @lpscr,

Fine-tuning only works for the supported languages. For non-supported languages it needs a couple of extra steps that are not implemented on the open-source repository.

[lpscr]

@Edresson thank you very much for the reply to know i try use with English method to replace the character Greek to English like i say in here #3229 and results its not very good , sometime it's good but most time not good, anyway , do you plan to support Greek ? i just wonder thank you very much for all your amazing work !

[erogol]

@lpscr we plan to support every language. But we need data. If you have some or are willing to help create some, come join our Discord and DM me.

For fine-tuning you can try transliteration Greek before passing to the model. So the available vocabulary would work that way.