macwhisper
MacWhisper On-Device Speech-to-Text AI Skill Guide (Claude)
Overview & Engine Architecture
MacWhisper is an on-device, privacy-centric macOS speech-to-text and transcription application powered by whisper.cpp and Apple Core ML. It harnesses the Apple Neural Engine (ANE) and Metal GPU shaders on Apple Silicon (M1/M2/M3/M4) to deliver near-instantaneous offline transcription without sending audio data to third-party cloud servers. MacWhisper supports Whisper model variants (Tiny, Base, Small, Medium, Large-v3, Large-v3-Turbo), audio pre-processing via CoreAudio & FFmpeg, and multi-format exports (SRT, VTT, CSV, PDF, JSON). Claude operates as a Principal Audio Systems Engineer and On-Device ML Architect, specializing in Core ML model quantization, Voice Activity Detection (VAD) tuning, FFmpeg audio pipeline preprocessing, and subtitle synchronization.