sciences

Deb Roy is the chief of the MIT Media Lab' S Cognitive Machines research group (!!) and the project on which he's currently working is at the very least unusual. The project is very serious and is called “The Human Speechome Project”. Speechome is the compression of speech and home. Good. ok.

The purpose is to understand how a human being acquires its language thanks to the social and physical context when he's a child. For doing this, Deb decided to record (almost) every moments of his baby's life, 9 mongths old. 400,000 hours audio and video during 3 years, from 12 to 14 hours a day.

Let's calculate: 3 years equal 15,000 hours. So there is redundancy: Roy installed 11 multi-directional video cameras and 14 microphones that will record every activity in the house. Hey hey, there is a button, called "ooops", that deletes the last minutes. True!
A disk of 5 terabytes stores the data temporarily before they're sent to Media Lab for analysis.

The analysis of this huge data flow, it is precisely the most interesting part of the project. To confirm their assumptions on the learning, the team of Roy must develop a software that's gonna take the place of his son, by analyzing the sights and the sounds perceived during these 3 years. Algorithms of speech or video processing are under development, for a better understanding of the all types of communication included in the images. New techniques of visualization already emerged from the project: distinguish a person who walks from one room to another, who drinks, who clean a baby...