FastScribe

Minimal, local audio transcription desktop app. Electron UI + a Python faster-whisper backend. Audio never leaves your machine — transcription runs fully offline after the model is downloaded once.

Download

Prebuilt installers for macOS (.dmg), Windows (.exe), and Linux (.AppImage) are attached to each release. They bundle the Python backend — no Python install required.

macOS: "FastScribe is damaged / malware" on first launch

The builds are not signed or notarized (that requires a paid Apple Developer account). Because the app bundles a Python backend, macOS Gatekeeper may quarantine it and — on recent macOS versions — even move it to the Bin, reporting it as malware. This is a false positive from the unsigned bundle, not actual malware.

To run it, strip the quarantine flag before first launch:

# 1. Open the .dmg and drag FastScribe to Applications, then:
xattr -cr /Applications/FastScribe.app
# 2. Open it normally.

If macOS already moved it to the Bin, restore it to Applications first, then run the command above. Alternatively, after a blocked launch, allow it under System Settings → Privacy & Security → Open Anyway.

Prefer to build it yourself and avoid this entirely? See Setup — locally built apps are not quarantined.

Features

• Drag-and-drop or file-picker for .mp3, .wav, .m4a, .opus
• Language selector (auto-detect or force a specific language)
• Voice-activity filtering to suppress silence hallucinations
• In-memory list of transcriptions with filename, timestamp, detected language
• Dark, minimal UI
• Python backend auto-spawned and cleanly terminated by Electron

Architecture

fastcribe/
├── main.js                 Electron main: spawns/kills backend, opens window
├── preload.js              contextBridge — exposes backend URL only
├── renderer/               UI (dark theme, dropzone, language selector)
│   ├── index.html
│   ├── styles.css
│   └── renderer.js
├── python_backend/
│   ├── main.py             FastAPI + faster-whisper (WhisperModel "small")
│   └── requirements.txt
└── package.json

• Backend: FastAPI on http://127.0.0.1:8000, POST /transcribe (multipart file + optional language) → { filename, transcript, language }.
• Lifecycle: main.js spawns the backend on launch (preferring the project virtualenv interpreter) and kills it on quit.

Prerequisites

• Node.js 18+
• Python 3.9+
• macOS / Windows / Linux

Setup

1. Python backend

cd python_backend
python3 -m venv venv
source venv/bin/activate          # Windows: venv\Scripts\activate
pip install -r requirements.txt
deactivate

The first transcription downloads the Whisper small model (~460 MB) and caches it; subsequent runs are fast.

2. Electron app

npm install
npm start

npm start auto-spawns the backend and opens the window once it's healthy.

Usage

1. Pick a language (or leave Auto-detect).
2. Drop an audio file onto the dropzone, or click Choose file.
3. The transcription appears at the top of the list.

Configuration

• Model size — edit WhisperModel("small") in python_backend/main.py. Larger models (medium, large-v3) are more accurate but slower.
• Port — backend runs on 127.0.0.1:8000 (see main.py and preload.js).

License

MIT