Setup That Matches the Hardware
Automated setup validates dependencies and prepares Whisper. Windows uses Vulkan on supported GPUs, while Apple Silicon uses Metal. This addresses installation and configuration work before the first transcription.
DESKTOP APPLICATION · LOCAL AI
EasyWhisperUI
Run Whisper locally without setting up a command-line workflow.
The Problem
People who want to transcribe recordings locally can run Whisper, but getting started involves model files, media conversion, dependencies, and hardware-specific configuration. Repeating those steps for a folder of recordings adds more work. I built EasyWhisperUI so users could transcribe locally without assembling that workflow themselves.
The Approach
EasyWhisperUI brings model selection, downloads, FFmpeg preprocessing, and transcription into one desktop app. Users can queue recordings, export text or timestamped subtitles, or use live transcription. Audio processing runs locally through whisper.cpp, with GPU acceleration where supported.
A Separate Boundary for Privileged Operations
The Electron renderer has no direct Node.js access. A narrow preload bridge connects the React interface to main-process operations such as file handling and transcription, keeping those responsibilities separate.
User-Requested Batch and Live Workflows
User requests shaped features including queue-based batch processing and live transcription. A sequential queue handles multiple recordings, FFmpeg converts inputs, and model downloads happen when needed. Live transcription handles ongoing audio and is currently marked beta.
Result and Compatibility
Reached 500+ GitHub stars as an open-source local transcription tool. Available for Windows, macOS, and Linux, with text and SRT output. Transcription speed depends on the model and available hardware; Linux compatibility can vary by distribution.