Updated faster-whisper to 0.10.0 with the ability to select a larger-v3 model.
Changed output folder to output folder inside the main folder.
If there is already a dataset in the output folder and you want to add new data, you can do so by simply adding new audio, what was there will not be processed again and the new data will be automatically added
Turn on VAD filter
After the dataset is created, a file is created that specifies the language of the dataset. This file is read before training so that the language always matches. It is convenient when you restart the interface
Fine-tuning XTTS Encoder
Added the ability to select the base model for XTTS, as well as when you re-training does not need to download the model again.
Added ability to select custom model as base model during training, which will allow finetune already finetune model.
Added possibility to get optimized version of the model for 1 click ( step 2.5, put optimized version in output folder).
You can choose whether to delete training folders after you have optimized the model
When you optimize the model, the example reference audio is moved to the output folder
Checking for correctness of the specified language and dataset language
Inference
Added possibility to customize infer settings during model checking.
Other
If you accidentally restart the interface during one of the steps, you can load data to additional buttons
Removed the display of logs as it was causing problems when restarted
The finished result is copied to the ready folder, these are fully finished files, you can move them anywhere and use them as a standard model
Added support for finetune Japanese
Changes in webui
1 - Data processing
2 - Fine-tuning XTTS Encoder
3 - Inference
Run Remotly
Run in Docker
docker
docker run -it --gpus all --pull always -p 7860:7860 --platform=linux/amd64 athomasson2/fine_tune_xtts:huggingface python app.py
Run Headless
bash
python headlessXttsTrain.py --input_audio speaker.wav --lang en --epochs 10 # Example with parameters
python headlessXttsTrain.py --help# See parameters