Skip to content

Church Cap v0.2.2 public preview

Choose a tag to compare

@Church-Cap Church-Cap released this 10 Jun 21:14
· 2 commits to main since this release

Church Cap v0.2.2 public preview

This update adds a new Performance section to the operator dashboard with a speed-vs-accuracy slider, Easy/Advanced settings, OS-aware Mac/Windows controls, live benchmarking, and offline recommended settings. Recommendations now choose conservative live-service defaults, while heavier models such as medium.en remain available for manual testing after benchmarking.
Windows GPU support has been tightened with better CUDA detection, cuDNN checks, CUDA troubleshooting controls, and a force reinstall flow for local NVIDIA runtime wheels inside .venv. The app now avoids recommending OpenAI Whisper/PyTorch CUDA paths on Windows when the built-in CUDA helper is intended for Faster Whisper/CTranslate2.

This release also improves live-caption stability by trimming obvious repeated word or phrase loops, adds individual QR code download buttons, adds a live performance tile, improves update-page wording, and refreshes documentation for the current Mac and Windows setup flow.

Highlights
• Added operator Performance panel with slider presets from fastest to most accurate.
• Added Easy and Advanced performance settings.
• Added Mac/Windows platform selection with auto-detection.
• Added Windows CUDA troubleshooting and force reinstall button.
• Added offline benchmark and live performance monitor.
• Added offline “Apply recommended” settings with safer live-service defaults.
• Added medium.en model option with latency warnings.
• Added individual QR code download buttons.
• Added performance status tile beside the live captions feed.
• Improved repeated word/phrase cleanup for live captions and transcripts.
• Improved Windows CUDA/runtime detection, including cuDNN DLL checks.
• Updated documentation, examples, and release hygiene for public GitHub publishing.

Notes
• Windows CUDA support uses Faster Whisper/CTranslate2 runtime packages, not PyTorch/OpenAI Whisper CUDA.
• Medium model is available, but should be benchmarked before live use.
• PowerShell scripts should be tested on Windows before tagging the final release, since they cannot be executed from the Mac-side development environment.