Sækja STT-CLI 2.0.0

v2.0.0 - Hybrid Speech-to-Text with Whisper Integration (November 26, 2025)

🚀 Major New Features:
- Offline Speech Recognition: OpenAI Whisper "tiny" model (MIT licensed) for fully offline operation
- Hybrid Engine Architecture: Auto-switch between Google (online) and Whisper (offline) based on connectivity
- Engine Selection Menu: System tray submenu to manually choose Auto-Detect, Whisper, or Google engines
- About Menu: New "About" section showing version, author, current engine, and Whisper status
- Batch Launchers: run.bat (normal mode) and run-debug.bat (troubleshooting mode) for easy startup

🔧 Technical Implementation:
- faster-whisper: 4x faster than vanilla Whisper with CTranslate2 backend
- INT8 Quantization: CPU-optimized for speed (greedy decoding, beam_size=1)
- Lazy Model Loading: Whisper loads on first use (~70MB one-time download)
- Model Cache: Stored in %APPDATA%\stt-cli\models for persistence
- Settings Persistence: Engine preference saved across sessions
- Hidden Imports: PyInstaller bundles av, faster_whisper, numpy, ctranslate2

⚡ Performance:
- Whisper Latency: ~1-2s transcription time (tiny model)
- Google Latency: ~0.5s transcription time (faster but requires internet)
- Model Size: 75MB (tiny model, one-time download)
- Executable Size: ~150MB (includes all Whisper dependencies)

🎯 Use Cases:
- Corporate Environments: Bypass Win+H restrictions with offline capability
- Air-Gapped Networks: Work completely offline with Whisper mode
- Privacy-Focused: Audio never leaves your machine in Whisper mode
- Variable Connectivity: Auto-switch between Google (accurate) and Whisper (offline)

📚 Documentation:
- Comprehensive architecture documentation in /docs
- Whisper model download and caching explained
- Threading model and flow diagrams
- Batch launcher usage guide

All v1.4.0 functionality remains intact:
- Auto-start on Windows Boot
- Double-tap Left Alt hotkey
- System tray operation
- Balloon notifications
- CLI window detection
- No admin rights required