Launches a server in localhost:5000 that transcribes audio files with faster_whisper. I recommend running it with docker to avoid dealing with cuda, venvs, and so on.
whisper-server uses the large-v2 version of whisper by default. If you want to use a smaller model, make sure you preload it in the Dockerfile and build it again.
localhost:5000/transcribelocalhost:5000/transcribe_segments
Build the docker image:
docker build -t whisper-server .Run it:
docker run --gpus all -it -p 5000:5000 whisper-serverTest it:
curl -X POST -F "file=@/path/input.wav" http://localhost:5000/transcribe