Skip to main content

Architecture

OpenTranscribe is built with a modern, scalable architecture.

System Components​

Frontend (Svelte)​

  • Progressive Web App
  • TypeScript
  • Responsive design
  • Real-time WebSocket updates

Backend (FastAPI)​

  • Async Python
  • RESTful API
  • OpenAPI documentation
  • WebSocket support

Workers (Celery)​

  • GPU Queue: Transcription, diarization
  • Download Queue: YouTube downloads
  • CPU Queue: Waveform generation
  • NLP Queue: LLM features
  • Utility Queue: Maintenance

Data Layer​

  • PostgreSQL: Relational data
  • MinIO: Object storage (S3-compatible)
  • OpenSearch: Full-text and vector search
  • Redis: Task queue and caching

Data Flow​

  1. User uploads file
  2. File stored in MinIO
  3. Celery task queued
  4. GPU worker processes:
    • Transcription (WhisperX)
    • Diarization (PyAnnote)
  5. Results stored in PostgreSQL
  6. Indexed in OpenSearch
  7. WebSocket notification to UI

Deployment Models​

  • Development: docker-compose with hot reload
  • Production: docker-compose with optimizations
  • Offline: Airgapped deployment
  • Cloud: AWS/GCP/Azure with GPU instances

Next Steps​