This project contains two scripts that use PocketSphinx to produce captions from a video file using speech recognition. The dependencies are a bit tricky, a Dockerfile is provided to produce a working environment. Specifically, the script currently relies on an unlanded patch to the PocketSphinx Gstreamer plugin.

caption.py takes a media file and generates a caption file. You can test this script with the pre-built docker image luser/auto-caption:0.2, for example:

docker run -t luser/auto-caption:0.1 ./run.sh https://people.mozilla.org/~tmielczarek/test-long.wav

Will produce captions on stdout.

adapt-from-captions.py takes a media file, a manually corrected captions file, and a PocketSphinx acoustic model, and adapts the model by feeding it the matched input audio and corrected text. It will output updated-model.tar.gz in the working directory if it succeeds.

Any copyright is dedicated to the Public Domain. http://creativecommons.org/publicdomain/zero/1.0/