Good Thinking

Silent headset lets users quietly commune with computers

Silent headset lets users quietly commune with computers
Arnav Kapur, a researcher in the Fluid Interfaces group at the MIT Media Lab, demonstrates the AlterEgo project
Arnav Kapur, a researcher in the Fluid Interfaces group at the MIT Media Lab, demonstrates the AlterEgo project
View 1 Image
Arnav Kapur, a researcher in the Fluid Interfaces group at the MIT Media Lab, demonstrates the AlterEgo project
Arnav Kapur, a researcher in the Fluid Interfaces group at the MIT Media Lab, demonstrates the AlterEgo project

Advances in voice recognition technology have seen it become a more viable form of computer interface, but it's not necessarily a quieter one. To prevent the click-clacking of keyboards being replaced by noisy man-machine conversations, MIT researchers are developing a new system called AlterEgo that allows people to talk to computers without speaking and listen to them without using their ears.

At first glance, the AlterEgo headpiece looks like the product of a design student who didn't pay attention in class. Instead of the familiar combination of an earpiece and microphone, the device is a cumbersome white plastic curve like the jawbone of some strange animal that hangs off the wearer's ear and arcs over to touch the chin.

It might look strange, but it's based on some fairly sophisticated technology. Inside the Alterego are electrodes that scan the jaw and face from neuromuscular signals produced when the wearer thinks about verbalizing words without actually speaking them aloud, a practice called subvocalization. This allows the headpiece to act as a microphone for a computer without actually picking up any sounds. Meanwhile, a pair of bone conduction headphones transmits sounds from the computer to the wearer that bypasses the usual channels and goes directly to the inner ear via the bones of the jaw and skull.

According to MIT, this makes communicating with a computer silent and completely private. One example of the benefits of such a system is is being able to use a computer as an aid to beating a chess opponent by silently communicating moves to the device and receiving advice surreptitiously. A more ethical use would be on the deck of an aircraft carrier, where it's normally too noisy to either speak or hear. But Arnav Kapur, leader of the project, sees a more fundamental application.

"The motivation for this was to build an IA device – an intelligence-augmentation device," says Kapur. "Our idea was: Could we have a computing platform that's more internal, that melds human and machine in some ways and that feels like an internal extension of our own cognition?"

In other words, such an interface could allow users to do away with passcodes, typing, and generally fiddling about with digital devices – distracting themselves and those around them from the business at hand.

Subvocalization has been known about for over a century, but when it comes to computer interfaces, it's still in its infancy. To figure out how to design the AlterEgo headset, the MIT team used an array of 16 electrodes set on various parts of a set of volunteers' faces to find the best neuromuscular signals. They were then asked to subvocalize a series of words four times in succession.

The team found that only four electrodes are actually needed, so they believe that a less cumbersome device is possible. Currently, the algorithms used operate on a vocabulary of about 20 words each for a series of computer tasks. These include multiplication problems and a chess application that allows moves to be entered using standard chess alphanumeric nomenclature.

Once the words are subvocalized, a neural network passes the data through a series of layered simple processing nodes that seek correlations between the neuromuscular signals and specific words. Usability tests on 10 subjects, who had spent 15 minutes tuning the prototype to their neurophysiology followed by 90 minutes used to carry out computer exercises, showed an accuracy of 92 percent.

However, Kapur says that this performance could be improved with more training data for the system. The hope is that by collecting more data, the system can go beyond simple commands and responses.

"We're in the middle of collecting data, and the results look nice," says Kapur. "I think we'll achieve full conversation some day."

The research was presented in a paper at this year's ACM Intelligent User Interface conference and the AlterEgo is demonstrated in the video below.

Source: MIT

Completely private until HAL learns how to read your lips. Seems like a simpler, less expensive alternative headset could be made with a laryngophone and earbuds or in-ear phones. Nobody would be able to hear your computer anyway and it would sound good enough to listen to music with, while I doubt the bone conduction phones would have good audio fidelity. You'd have to whisper with the laryngophone, but that's still relatively quiet and no neural networks would be needed.
I like it. I subvocalize. If this thing worked 100% then I would consider getting the 4 sensors implanted. Pretty sure I can subvocalize quicker than I can type.
I'm not interested if it has to be sent to google and then they figure out what I said. Because I subvocalize a lot of my thoughts and don't fancy them all running through the cloud.