XScribe

A cross-platform, local-first transcription application with speaker diarization. Think Otter.ai, but open-source and running entirely on your device.

Features

• Local-First: All processing happens on your device - no data leaves your computer
• Speaker Diarization: Automatically identifies and labels different speakers
• Cross-Platform: Works on macOS, Windows, and Linux
• Multiple Languages: Supports 99+ languages via Whisper
• Privacy Focused: No accounts, no telemetry, no cloud processing

Quick Start

1. Install Dependencies

npm install

2. Download Models

# Download tiny.en Whisper model + diarization models (~100MB total)
npm run download:models

# Or choose a larger model for better accuracy:
npm run download:models:base    # ~150MB - good balance
npm run download:models:small   # ~450MB - better accuracy

3. Run the App

npm run dev

4. Build for Distribution

# Build for current platform
npm run dist

# Build for specific platform
npm run dist:mac
npm run dist:win
npm run dist:linux

Technology Stack

• Framework: Electron + Vite + Vue 3 + TypeScript
• Speech Recognition: sherpa-onnx with Whisper models (WASM-based, cross-platform)
• Speaker Diarization: pyannote segmentation + wespeaker embeddings via sherpa-onnx
• Build: electron-builder for cross-platform distribution

Model Information

Whisper Models (Transcription)

All models use sherpa-onnx's ONNX-converted Whisper format:

Model	Size	Languages	Quality
tiny.en	~60MB	English only	Basic
base.en	~120MB	English only	Good
small.en	~400MB	English only	Better

Diarization Models (Speaker Identification)

• sherpa-onnx-pyannote-segmentation-3-0.onnx - Speaker segmentation
• wespeaker_en_voxceleb_resnet34.onnx - Speaker embeddings

Both are downloaded automatically by the download:models script.

System Requirements

• OS: macOS 10.15+, Windows 10+, Ubuntu 20.04+
• RAM: 4GB minimum (8GB+ recommended)
• Storage: 200MB - 1GB depending on model choice

Project Structure

xscribe/
├── src/
│   ├── main/           # Electron main process
│   │   ├── index.ts    # App entry point, IPC handlers
│   │   └── services/   # Transcription & diarization services
│   ├── preload/        # Secure IPC bridge
│   └── renderer/       # Vue frontend
│       ├── App.vue
│       └── components/
├── models/             # ML models (gitignored)
├── scripts/            # Helper scripts
└── build/              # Build resources

Development

# Install dependencies
npm install

# Download models
npm run download:models

# Start development server with hot reload
npm run dev

# Type check
npm run typecheck

# Build production
npm run build

Privacy

XScribe is designed with privacy as a core principle:

• No cloud processing: All transcription and speaker identification runs locally
• No telemetry: We don't collect any usage data
• No accounts: No sign-up or login required
• Your data stays yours: Audio files are never uploaded anywhere

License

MIT License - See LICENSE for details.

Acknowledgments

• OpenAI Whisper - Speech recognition model
• sherpa-onnx - ONNX inference runtime
• pyannote-audio - Speaker diarization models