KB Whisper Large API

FastAPI wrapper that exposes OpenAI-style transcription endpoints backed by Whisper-compatible models from Hugging Face.

Endpoints

• GET /
• GET /health
• GET /v1/models
• POST /v1/audio/transcriptions (OpenAI-compatible multipart form)
• POST /transcribe (simple alias)

OpenAI-style request example

curl -X POST http://localhost:8000/v1/audio/transcriptions \
  -F "model=KBLab/kb-whisper-large" \
  -F "file=@./sample.wav" \
  -F "language=sv" \
  -F "response_format=json"

model is selected per request. On first use, the model is downloaded from Hugging Face and cached under HF_HOME.

Environment Variables

• WHISPER_MODEL_ID (default: KBLab/kb-whisper-large)
• WHISPER_MAX_MODELS_IN_MEMORY (default: 2)
• WHISPER_DEVICE (auto|cpu|cuda|mps, default: cuda in container)
• WHISPER_DEFAULT_LANGUAGE (default: sv)
• WHISPER_MAX_UPLOAD_SIZE_MB (default: 200)
• WHISPER_ENABLE_DOCS (true|false, default: true)
• HF_TOKEN and HF_ENDPOINT are supported through Hugging Face libraries.

Runtime model behavior

• If model is omitted, the API uses WHISPER_MODEL_ID.
• Models are loaded lazily on first request.
• The API keeps up to WHISPER_MAX_MODELS_IN_MEMORY active models in an LRU cache.
• If a model cannot be loaded, the API returns 400 invalid_model.