When Machines Listen

Researchers are training computers how to listen - and to distinguish, say, the sound of breaking glass from the squeak of a rubber toy.

By Mary Jane Irwin, Business 2.0 Magazine

(Business 2.0 Magazine) -- Never mind that cameras that watch your every step. The future is all about microphones that eavesdrop on your life, according to researchers in the growing field of machine listening.

Sure, intelligent software can examine hours of security video footage and, say, try to recognize intruders, but it's far easier to train a computer to listen for the sound of breaking glass. And there are plenty more places where the automatic identification of strange noises can be useful.

"People aren't aware of how much you do using your ears," says Paris Smaragdis, a research scientist at Mitsubishi Electric Research Laboratories. Smaragdis has spent three years looking at industrial applications for machine listening, creating software that can track movement audibly and distinguish the ring of a phone from the squeak of a rubber toy.

Right now, audio-comprehension software is focused on speech recognition, but Smaragdis is betting that his technology will one day supplant visual imaging. For one thing, it's cheaper to capture and process audio than video. "A high-quality camera can cost $500," he says. "A high-quality microphone can cost $50."

Audio files also eat up less bandwidth, saving more money. "We're ready for prime time," Smaragdis says. "It's a matter of people grasping the technology." So perk up your ears: Here's his primer on the future applications of machine listening.


It can cost companies up to $200 an hour to recover data from a fritzed hard drive. But tiny microphones planted in a computer can monitor its hard drive for wear and tear--and send a message warning the user to back up data before the PC goes nuclear.


A guard at a bank of screens can't always monitor the right one. Listening algorithms can filter out the air-conditioning and hear a door opening in an abandoned corridor. By triangulating the sounds of footsteps, the software can then tell security cameras where to point.


Listening devices can monitor how efficiently office space is put to use. If a spare conference room is filled with hubbub less than once a week, it might be time to turn it into a cube farm.


When vehicles break down, as long-suffering drivers know, they tend to make a racket. By analyzing the clanks and creaks of an auto's carriage, an embedded listening device can tell you when to buy a new muffler or replace your brake pads.


Your workers shouldn't be wasting an entire afternoon watching the ball game. An intelligent video drive can create an instant highlights reel by listening for the sounds of cheering and showing only the minute before and the minute after. Top of page

To send a letter to the editor about this story, click here.