SPACE_README.md

Live Translation App

This Space serves both the frontend (React + Vite build) and backend (FastAPI) from a single Docker container.

How to use

• Open the Space URL. The UI loads from / and the API is same-origin.
• Host view:
  • Start Recording for sentence-level live transcription
  • Upload audio for server-side transcription (use the Language hint for accuracy)
  • Click "Preload Model" to warm the translator

Endpoints

• GET /health → { "status": "ok" }
• POST /translate → { text, source_language, target_language }
• POST /preload → { source_language, target_language }
• GET /metrics → Prometheus metrics

Space variables (recommended)

• STT_ENABLED=1 – enable server-side transcription
• WHISPER_MODEL=small – good for CPU (use medium for better accuracy; large-v3 requires GPU)
• Optional: REQUESTS_PER_MINUTE=60, ALLOWED_ORIGINS=*

Notes

• First requests may download models. Use Preload to warm up.
• For best accuracy on fast or unclear audio, set WHISPER_MODEL=medium (CPU) or upgrade to a GPU Space and use large-v3.