d3vv3 / briefcast

briefcast

Listen to the most important parts of your favorite podcasts, in the time you have available.

The pipeline

whisper-api

The whisper-api is a RESTful API that provides a way to interface with faster-whisper via API. It uses fastapi and is just a passthrough to the faster-whisper python transcribe library.

Just send the audio and the API will return the transcribed text as a text file.

curl \
  -i -X POST "localhost:11000/transcribe?beam_size=5" \
  -F audio_file=@some_podcast.opus

You can add faster-whisper transcribe arguments as query parameters.

Any audio file will work, since faster-whisper uses ffmpeg to convert the audio to the correct format.

The image is CUDA enabled, so it will use the GPU if available.

chopper

The chopper is a simple fastapi API that takes a whisper transcript file with dropped (removed) lines, and the original audio file, and returns a new audio file trimmed accurately to the input transcript file.

curl \
  -i -X POST "localhost:11001/chop" \
  -F transcription_file=@/tmp/chopped_transcript.txt \
  -F audio_file=@/tmp/some_file.opus \
  -o result.opus

Deployment

briefcast comes with a docker-compose.yaml, that enables you to deploy the whole pipeline easily.