MIT eggheads craft creepy covert speech-to-text interface
Boffins at MIT have developed a silent speech interface called AlterEgo that allows wearers to communicate via text translated from neuromuscular signals – tiny face twitches – without a visible tell.
It’s a mechanism for covert communication between people and machines. Well, covert to a point: you have to wear a claw-like gadget on your fizog for it to work.
The device is described in a paper titled, “AlterEgo: A Personalized Wearable Silent Speech Interface,” presented last month at the Association for Computing Machinery’s ACM Intelligent User Interface conference.
AlterEgo, a mask that resembles a cross between dental headgear and Borg prosthetic, is something short of a thought-to-text system; it listens for intended but not vocalized speech, which is to say the physiological response arising from talking to oneself without visible lip or facial movements.
“A user’s intention to speak and internal speech is characterized by neuromuscular signals in internal speech articulators that are captured by the AlterEgo system to reconstruct this speech,” the paper explains.
Muttering to oneself, in other words, is measurable.
Silent speech recognition has been demonstrated by sticking implants into the brain, placing sensors on the tongue, and other means of monitoring movements and physiological signals.
But as MIT Media Lab researchers Arnav Kapur, Shreyas Kapur, and Pattie Maes point out, these approaches have the obvious drawback of being more than a little invasive and impractical.
There have been non-invasive experiments along these lines, to decode language from lip movements and from facial muscles movements, but these approaches haven’t proven particularly accurate.
“The key difference between our system and existing approaches is that our system performs robustly even when the user does not open their mouth, make any sound and without the need for any deliberate and coded muscle articulation that is often used when using surface EMG to detect silent speech,” the researchers state in their paper.
On a standard digit recognition test, the researchers claim that AlterEgo performs with 92 per cent accuracy, better than other approaches and without the need for lip reading tech. They suggest the system performs comparably to state-of-the-art speech recognition systems, though with a more limited vocabulary.
The boffins describe a variety of mundane applications for the technology: math calculations requested in silence and returned from a machine; controlling IoT devices and media playback; covert calendaring; and fielding phone calls without a sound.
Perhaps too, the tech could find a way to give people with certain disabilities another communication option or to assist in police or military operations where silent communication would be useful.
And if there were a way to mask the mask itself, it could become a game-changing tool for card cheats. ®