Abdoul2146/english_to_hausa_backend

# Hausa Translator Backend API (v2.0) A production-grade, highly scalable, and fully decoupled FastAPI backend featuring three modular processing pipelines. This backend is optimized to run on GPU-enabled hosting environments (such as a Lightning AI GPU Studio) while persisting transaction and progress metadata to a managed Supabase PostgreSQL instance. --- ## 🚀 Key Features * **Clean Separation of Concerns:** Rigid domain layering (`models`, `schema`, `crud`, `services`, and `api/endpoints`) ensures maximum maintainability and code clarity. * **Fully Decoupled Architecture:** Users can chain outputs or run individual pipeline stages completely standalone. * **No In-Memory Dependencies:** Built on PostgreSQL using SQLAlchemy and Alembic, allowing infinite application scaling and seamless recovery on restarts. * **Decoupled Audio Extraction:** Stage 1 is fully decoupled. You can download and extract audio from public URLs (YouTube, Vimeo, etc.) OR upload raw video files directly to the server. * **Interactive Transcription & Translation Flow:** Stage 2 is split into distinct, modular operations. Transcribe English audio to English text, review or copy it, and then submit it for segment-by-segment Hausa translation preserving all timestamps. * **Robust Security:** Endpoint operations are protected via custom `X-API-KEY` authorization headers to prevent GPU compute resource exhaustion. * **Lazy Singleton Model Cache:** Deep learning models (Whisper, NLLB-200, MMS-TTS) load lazily on demand and stay cached efficiently in GPU memory. --- ## 📂 Codebase Directory Layout ``` english_to_hausa/ ├── api/ │ ├── endpoints/ │ │ ├── video_to_audio.py # URL extraction & Direct file upload pipelines │ │ ├── translate.py # Whisper transcription & NLLB translation pipelines │ │ └── tts.py # MMS-TTS synthesis pipeline │ └── deps.py # API Key security validation & DB dependency injection ├── crud/ │ └── job …