More than 80% of all data in a company is unstructured information. This includes emails, word documents, paper documents, images, web pages, videos and hundreds of additional formats. Unfortunately, attempts to use this immense and strategic resource often fail because many businesses lack the requisite technology to understand and effectively utilize content that resides outside of traditional structured databases.

The question is: what to do to understand this unstructured information? And the solution is offered by automatic indexation! Or, in other words, how to structure unstructured data.

FRANCE24, uses some of the audiovisual indexation tools of Autonomy - a leading company in automatic indexation - for its archive system. Thanks to these tools, through video and audio analysis, documentary files are automatically filled out as soon as they are created. Then FRANCE24 archivists complete the files. The goal is to provide users with as much information as possible before any human intervention!

The automatic indexation used in FRANCE24 consists of several technological solutions:

• Speech recognition: each video’s audio is analyzed, extracted, and indexed with the TimeCode. Even if this “speech to text” functionality is not perfect, it allows the extraction of some of the video’s concepts without any human intervention.

• The Automatic Scene Detection automatically generates a storyboard, that is to say a succession of shots that represent the video. Each time a transition is recognized a shot is captured. Just in a glance, users know what the video is about.

• Letter recognition: the system analyzes each image in order to detect letters or words inside the videos. So you can automatically catch the name of a guest in a program that appears on screen.

So that is what FRANCE 24 uses, but you can add many other functionalities:

• Speaker tone recognition: first we train the system with the tone of the people speaking who we want to be recognized. Then the system can analyze the videos and be able to tell which speaker is talking when in the video! Great functionality when, in a debate, you search for some specific sentences of one of the panelists.
• Face recognition: using the same method, we train the system with the faces of the people we want to be recognized in the videos. Then the system can analyze the videos and tell who are the people in the video and when they appear!