Accelerate Networks Transcription Service

a little REST API to perform audio transcriptions

This is a small python tool to transcribe audio asynchronously.

Install

1. Install whisper
2. Install the application's requirements: pip3 install -r requirements.txt
3. for production deployments, install gunicorn: pip3 install gunicorn
4. Install redis: sudo apt install -y redis-server or use a container: podman run -p 6379:6379 docker.io/library/redis

Once all dependencies are installed, run the web server: gunicorn -b 0.0.0.0:5000 an_transcriptions:app and start the backround worker: rq worker. The worker must share a filesystem (at least /tmp) with the web server.

Authentication

When started, the program will attempt to read a file api_keys.txt from the current working directory. Each line of this file will be treated as an API key, up to the first comma character. For example:

API_KEY_1, assigned to user a
API_KEY_2, assigned to user b

here API_KEY_1 and API_KEY_2 will be accepted as valid API keys.

Endpoints

• POST /enqueue - request a transcoding
  • Upload the file with form field file: curl -X POST http://transcription-service:5000/enqueue -H "Authorization: Bearer API_KEY_1" -F file=@/tmp/transcode-me.wav
  • You will get a response with a job ID back: {"id":"f757e53e-bad8-402a-94e7-ccc11baf39c5"}
• GET /j/<job_id> - check on the status of a transcode request
  • response will be JSON with a status and possible result: {"status": "finished", "result": "the transcribed text"}
  • status may be one of queued, started, deferred, finished, stopped, scheduled, canceled and failed
  • result will be null unless status is finished.
  • the job results will be purged after 500 seconds (a little over 8 minutes)