TORONTO, ONTARIO - Videos are a predominant part of the modern internet experience. Hundreds of millions of hours of content are readily available on various online platforms, most notably YouTube. One can easily find content on almost any topic imaginable, but curating it is far more arduous. This project describes the creation of Laconic, a deep learning-based extractive video summarization tool that extracts the essential parts of any video so users can then focus on the most important segments.

Origins, purpose of Laconic 

Attention spans are shrinking worldwide - especially among youngsters - even as students are expected to learn a seemingly ever-expanding amount of information.¹ With video platforms like YouTube uploading hundreds of hours of video content per minute, one has access to a plethora of video resources from which to educate oneself. This is often too much of a good thing: With so much content, how can one efficiently locate an object sought? One of the answers comes from automatic video summarization - which is the objective of this project, dubbed ‘Laconic.’ 

Laconic generates summaries of longer videos by selecting key elements and most interesting materials. Generating a transcript of an original video and using a machine learning model to identify the key elements in the text and create multiple clips of these achieves this. The output is composed of a set of video clips extracted from the original video and edited together into one shorter version of the original. By watching a summary by Laconic, users can digest the contents of a long video in just a few minutes and determine the usefulness of the video to decide whether to watch it in its entirety. 

This report begins by defining what Laconic is and what it comprises. Then, it provides a detailed description of all the com