Synthalingua Update Log

Core Application (transcribe_audio):

• Improved Output Filtering and Formatting: The core transcription logic now includes enhanced filtering of blacklisted words and phrases using regular expressions. This filtering is applied consistently to both the console output and the data sent to the web server for display. The output formatting has been standardized, with clear section headers for different kinds of output (original, translation, transcription). A more concise, one-line format is used when the --no_log flag is active.
• Automatic Model Swapping Removed: The --auto_model_swap flag and its associated logic have been removed. This feature is now considered deprecated.
• Improved handling of 12gb-v3 model and Mel Calculation: The script now uses 128 mels when loading the 12gb-v3 model, enhancing accuracy.
• Fine-tuned Model Handling (Obsolete): The code related to loading fine-tuned models is marked as obsolete, and the --use_finetune flag is temporarily non-functional due to model retraining.
• Minor Performance Optimization: The sleep duration in the main loop has been reduced.

Modules:

• Subtitles Generator (sub_gen.py): The script now uses arguments for device selection and model directory when generating subtitles.

Updater:

• Updater Created: You can now download a small updater per major update to skip redownloading the big zip file every major. This will only work per previous version in this case, the updater will patch 4.62 to .63.