Transcription Engine

Harvey provides two distinct engines for local transcription. You can switch between them in the Configure tab under the Transcription section.

Note: Only models belonging to the currently selected engine will be available for transcription. If you switch engines, you must download at least one compatible model for that engine.

1. Whisper.cpp

• Recommended for macOS: Highly optimized for Apple Silicon (M1, M2, etc.) using Core ML.
• Performance: Extremely fast and efficient on Mac hardware.

2. Faster-Whisper

• Recommended for Windows (CPU): Uses the CTranslate2 backend, which is often significantly faster than whisper.cpp on Intel and AMD CPUs.
• Performance: Provides an excellent balance of speed and accuracy for systems without high-end GPUs.

Managing Models

For either engine, you can download various model sizes (Tiny, Base, Small, Medium, Large). Larger models provide higher accuracy but require more RAM and processing power.