Box taps machine learning to pull data from videos, images and audio

Brittany A. Roston - Oct 11, 2017
Box taps machine learning to pull data from videos, images and audio

Box has announced its new ‘Box Skills,’ a framework that taps machine learning to power new features that extract data from multimedia files. With these Skills, the system is able to automatically do things like extract text from images, create a text transcript from an audio file, and recognize faces within video files, among other things.

Box announced the new framework today, pointing toward the increasingly huge amounts of data that are being created every year. Manual processing all of that data is a huge burden, one that is increasingly more expensive for companies and time-consuming for everyone else. This results in a mass of data that can’t be properly or efficiently utilized, but machine intelligence promises to solve that problem by doing the hard work itself.

Box Skills brings machine learning to the Box storage service, and it is behind three abilities that will soon be available to content stored on the service. The first among them is image intelligence, a feature that has been in beta testing since this past August. Images are the second most common type of content stored in Box, the company says, and with this technology it is easier than ever to manage.

The image intelligence works by automatically adding context to images via detecting things about them, such as text that it may contain, or handwriting when applicable. The text is automatically added as part of the image’s metadata, and tags are applied as well so that it is easier to find images when necessary.

The other skills are similar to this, such as audio intelligence’s ability to create a text transcript of an audio file. The skill is also able to figure out the topics of an audio file ultimately making it easier to find needed audio clips by searching for specific topics or words they contain.

Then there’s video intelligence, which is able to recognize faces within footage, as well as figure out the video’s topic(s) and transcribe any audio in them. Users can leverage this information to more quickly navigate through a video in search of a particular segment, as well as finding videos while searching through the Box files.

The Box Skills will enter beta early next year.

SOURCE: Box Blog

Must Read Bits & Bytes