Background#
Right now my NAS handles audio-to-text conversion, but without a GPU, anything more than a short clip takes hours to process. So I exposed the Whisper instance running on a desktop with a 4060 as an API service instead, and pointed the NAS at it.
Wrapping it#
Wrote a small FastAPI app that wraps the Whisper model behind an API endpoint capable of accepting an externally-uploaded audio file — loading the medium model and handling transcription requests through a /transcribe/ POST endpoint.
Since it’s only reachable from inside the home network, I skipped token authentication for now.
