Repository navigation
What's New
AI Transcript Enhancement
- Local LLM integration via LM Studio for post-processing transcripts
- Screen context awareness for smarter enhancement
- Formatting toggle and thinking tag handling
Whisper Fine-Tuning Pipeline
- Complete LoRA fine-tuning pipeline for programming vocabulary
- WER improved from 21.8% to 1.5% on programming terms
- CoreML export with quantized variant for Neural Engine
- Custom ProgrammingGlossary and TechnicalDictionary for prompt tokens
Audio Safety Net
- Rotating WAV buffer saves recordings automatically
- History view with playback and re-transcription
Bug Fixes
- Fixed transcription truncation (gap assembly, hallucination filter, tail threshold)
- Fixed model reload, device observer, audio resampling issues
- Fixed clipboard handling and cancel behavior
- Improved VAD sensitivity and silence timeout
New Services
- AudioRecordingService — persistent recording buffer
- LMStudioService — local LLM communication
- ScreenContextService — frontmost app detection
- TranscriptEnhancementService — AI post-processing orchestration
UI Improvements
- Expanded settings view with model selection, formatting toggle, sidebar settings
- Rich history view with timestamps and status
Documentation
- Contributing guidelines (CONTRIBUTING.md)
Install
Download VoiceNative.dmg below, open it, drag to Applications. Requires macOS 15+ on Apple Silicon.