Tools for extractive summarization based on lexical and prosodic features for the inEvent Project. Most of the code is written in R with bash shell script wrappers.

Requires:

• praat (for prosodic features)
• R (3.1.1 or higher)
• python: Theano (for MLP)

The main wrapper is ineventwrappers/inevent-summarization.sh which at the moment takes as input:

• ASR output in JSON format (Augmented ASR in the inEvent API)
• A .wav file
• A media file (e.g. .mp4)
• The working directory It returns a JSON file of time aligned quotes from with extractive summary scores (probabilities).

Note a large amount of extra feature files and debug information is currently produced. More documentation will appear in the future.

Notes:

• Although the relevant function is called extract-spurt-feats.sh, we could be using any sort of segment really with a start and endtime. For this system, this is output after processing ASR json files.