v1.6.6.1

This is a demo of a different whisper library, it has not been thoroughly tested so there are likely bugs

v1.6.6.1 (current)

• more robust output purification and removal
• metrics on by default for this version
• metrics changed to be more visible and distinguishable
• fixed model download loop for models that were considered too small to be real

v1.6.6

• Fixed memory leak in previous version (which caused crashing, oopsie)
• Min VAD Duration (s) and Min Avg Confidence have been added to help fight against hallucinated outputs, you can enable debug text to see everything that gets filtered out

v1.6.5.9

• Whisper Library Changed from https://github.com/Const-me/Whisper to https://github.com/sandrohanea/whisper.net (so complete rework behind the scenes)
• Whisper can be "Recognize Once" or continuous
• Whisper model selection now uses absolute path
• Whisper can be GPU (and CPU) or just CPU
• Whisper now supports most if not all models here https://huggingface.co/ggerganov/whisper.cpp/tree/main (including quantized models)
• Whisper now supports distil models: https://huggingface.co/collections/distil-whisper/distil-whisper-models-65411987e6727569748d2eb6