Voice Mode

# Core ML Implementation Summary ## Changes Implemented ✅ ### 1. Core ML Download Function **File**: `voice_mode/utils/services/whisper_helpers.py` - ✅ Added `download_coreml_model()` - Downloads pre-built models from HuggingFace - ✅ Modified `download_whisper_model()` to use pre-built models on Apple Silicon - ✅ Deprecated `convert_to_coreml()` for backward compatibility - ✅ Removed dependency on `coreml_setup.py` ### 2. MCP Tool Updates **File**: `voice_mode/tools/whisper/model_install.py` - ✅ Removed `install_torch` parameter completely - ✅ Removed `_handle_coreml_dependencies()` function - ✅ Removed all PyTorch/coremltools installation code - ✅ Updated status messages for pre-built models ### 3. Platform Detection - ✅ Checks `platform.system() == "Darwin"` and `platform.machine() == "arm64"` - ✅ Only attempts Core ML download on Apple Silicon - ✅ Falls back to Metal acceleration if Core ML unavailable ### 4. Testing - ✅ Successfully tested download of Core ML model - ✅ Verified file structure and extraction - ✅ Confirmed 15.7MB download for tiny model ## What Still Needs Work ⚠️ ### 1. Whisper Install Default Model **File**: `voice_mode/tools/whisper/install.py` - ❌ Does NOT currently install a default model - ❌ Needs `--model` and `--no-model` flags added - Current: Just installs whisper.cpp binary - Needed: Should install base model by default ### 2. Documentation Updates - ❌ Remove references to Xcode requirements - ❌ Remove references to PyTorch installation - ❌ Update README with new simplified process ### 3. Potential Homebrew Integration **Discovery**: Official `brew install whisper-cpp` exists! - Could potentially use Homebrew formula instead of building - Includes Metal acceleration support - Would eliminate compilation entirely - Consider for Phase 2 improvement ## Architecture Note Both CLI and MCP tools use the same underlying functions: - `whisper_model_install` (MCP tool) → calls `download_whisper_model` - CLI commands → also call `download_whisper_model` - This ensures consistent behavior across interfaces ## Next Steps 1. **Phase 1b**: Update `whisper_install` to include default model 2. **Phase 2**: Consider using Homebrew whisper-cpp formula 3. **Phase 3**: Clean up legacy code and documentation 4. **Phase 4**: Remove coreml_setup.py entirely ## Performance Impact - Pre-built Core ML models: 2-3x faster than Metal - No build time required (instant vs minutes) - No 2.5GB PyTorch dependency - No Xcode installation required - Works on all Apple Silicon Macs (M1/M2/M3/M4)