Tuesday, August 22, 2006

Neural Net Makes Eavesdropping Breakthrough

Kevin Poulsen writes on 27B Stroke 6:

Government-funded researchers have discovered a mathematical solution for separating a single voice from a multitude of other sounds and voices -- solving what scientists call the "cocktail party problem," and the rest of us know as Harry Caul's plight in The Conversation.

In research funded by the National Science Foundation and -- big surprise -- the National Security Agency, two math professors from the University of Missouri-Columbia and one from Siemens trained a neural network to reconstruct a voice in a crowd with greater accuracy than existing techniques.

More here.

0 Comments:

Post a Comment

<< Home