Whisper Lambda

This repo hosts the code for the Whisper Lambda, which is a Lambda function that listens on API calls to transcribe audio files.

Usage

WARNING: Has a dependency on ffmpeg, which is not installed in the lambda.

Pull the model dependencies:

make pull

Run the app locally:

make run

# In another terminal
echo "{ \"model\": \"base\", \"file\": \"$(base64 -i ./sample_1min.mp3)\" }" > req.json
curl -v -X POST \
  'http://127.0.0.1:9001/lambda-url/_/' \
  -H 'content-type: application/json' \
  -d @req.json

NOTE: Before deployment, remove all the extra files from the model directory, to avoid huge lambdas.

Deployment

TBD

TODO

• Fix the ffmpeg dependency
• Add deployment instructions
• Add terrafom scripts
• Add CDK scripts
• Add SAM scripts
• Add CI/CD to this repo