When I met with Starner in November, he was constantly grabbing for the Twiddler, jotting down a note, and then putting it back. It was all very unobtrusive – as long I kept my eyes on his face, rather than on his hands. But what about for people who don’t want to learn how to use a Twiddler, or don’t want to walk around with a four-inch piece of Velcro stuck to an ever-present fannypack that’s crammed full of electronics?
One possibility, says Starner, is what he calls “dual-purpose speech.” The idea is to combine a voice recognition system with people’s tendency to repeat words in conversations to trigger responses on the part of their own wearable computers.
Let’s say I’m walking down a hallway and ask Starner if he wants to meet with me next week.
“Next week?” he might say in response. I don’t know it, but just before Starner said “next week,” he pressed a little push-to-talk button, perhaps in his pocket. The computer heard the phrase “next week” which is a voice command that brings up next week’s calendar in his head-mounted display.
“How about Monday?” I ask.
“Monday,” he says, pressing the button again. That’s another command, of course. The calendar flashes to the Monday view. “I’m busy all day,” he says. “How about Tuesday?”
By choosing which of the spoken words the computer actually processes, it’s possible to use dual-purpose speech to navigate through and update an electronic day planner. Although speech recognition is still pretty rough for many applications, this sort of low-vocabulary recognition is completely within the capabilities of today’s systems.
Even better is that these voice recognition patterns are the norm when people speak to each other, which makes the dual-purpose speech solution a potential boon when coupled with a push-to-talk button. This was apparent from one happy interaction between Starner and former student Ben Wong, who had a habit of leaving his microphone open and running a full-text speech recognition system on the mic’s input.
On that day, the two were discussing interesting readings when Wong suggested a particular article of interest. Most of what was on the student’s screen was gibberish since full-text systems aren’t very good at transcribing conversational speech. But suddenly Wong stopped and repeated the article’s title slowly and clearly, then resumed his normal speech pattern. The system locked right on and got the title and reference perfectly. Even better, if Starner hadn’t been primed, he wouldn’t have thought anything of the incident since the repetition seemed completely normal.
Voice input is pretty neat, but for speed, accuracy and flexibility it’s still no match for the Twiddler. Although it looks intimidating and hard to use, another paper that Starner and his students published last year found that people can learn it more quickly than they can a QWERTY keyboard, and they can type upwards of 60 words per minute with just a few hours of practice.
Starner goes on to say in the paper that one potential breakthrough application for the Twiddler is to help the deaf, who “have adopted wireless texting as a convenient means of communication within the community.” With a Twiddler connected to a cell phone texting on a handheld mobile device could rapidly become an almost natural experience.
Meanwhile, I’m going to get my own Twiddler. If I can master the art of one-handed typing, I might just become a cyborg myself.