Swap the string "True" / "False" to boolean before passing as they always return True in if statement and does not properly disable the functions even when set to disable in webui.
Interface cleaned
Documentation improved & expanded
Dataset generation improved:
- Better VAD detection
- Better debugging/information
- Better editing/managing transcriptions
- Basic audio editing
- Re-transcription on detections of duplicate audio when splitting large files.
- Control over model precision fp16/fp32/mixed
- Migrate to openai-whisper (also saving 1.5GB on installation requirements for cublas)
Testing stage allows you to choose what minimum length audio
Model export exports all audio with text file descriptions of too long/too short
Misc other improvements too numerous to mention
Close GPTTrainerConfig( on line 965
analytics_enabled=False for Gradio (stop it complaining about new versions of Gradio)
demo.queue().launch( line 2387 to launch Gradio.
Updated the numbering of generated wav's so they match within the CSV files.
Added a dataset validation option to compare the Whisper generated wav files vs the transcribed text in the CSV files. This is purely to point you in the direction of data that may need changing/editing within your training dataset.
TTS Generator - Various interface bugs & filtering options cleaned up.
TTS Generator - TTSDiff now scans generated text and TTS for errors.
TTS Generator - TTSSRT now creates subtitle files for video production e.g. a Youtube video.
Finetune - Now uses a customised tokenizer to deal with Japanese.
Finetune - Pre flight check and warning messages.
Finetune - Extra documentation and warnings.
Entire file structure has been re-organised to simplify management and future changes.
Documentation (built in and Github) has been rewritten/tidied up.
Requirements files have been cleaned up and simplified.
ATsetup has been re-written as necessary with additional options.
DeepSpeed moved up to version 14.
Standalone Application moved to PyTorch 2.2.1.
Nvidia CUDA Toolkit installation is NO LONGER needed (other than to compile DeepSpeed on Linux)
Tested on Linux and Windows.
Improved some areas of the documentation to clarify some issues.
Dropped worker threads down ONLY when Japanese is used, due to limitations of the translation.
j.katzwinkel@posteo.de