NASA is preparing to test a voice-recognition computer system designed to make life easier for astronauts--unlike the murderous but fictional HAL.
Called Clarissa, the system was developed to help astronauts on the International Space Station with routine tasks--such as testing the water supply for contamination--by responding to verbal commands and speaking detailed instructions. That way, technicians can concentrate on the task, rather than scrolling through computer screens, flipping manual pages, or possibly, reining in a mouse floating overhead.
Future applications of the system, if it's successful, could include helping astronauts during space walks or with research missions outside of the capsule.
"As far as we know, this is the first ever spoken-dialogue system in orbit," said Manny Rayner, Clarissa lead implementer at NASA's Ames Research Center.
NASA is demonstrating Clarissa on Friday during a meeting of the Association for Computational Linguists, at the University of Michigan, Ann Arbor. By next week, astronaut John Phillips is expected to complete training staff aboard the International Space Station, in preparation for later use of Clarissa.
Designed over three years by a team of scientists at the Ames Research Center, Clarissa is a "virtual crew assistant" based on a voice-recognition platform from Nuance. A PC that responds to roughly 300 voice commands, Clarissa will read procedures or retrieve other information.
The help is much needed--astronauts at the space station have roughly 12,000 different procedures associated with their jobs. Those include checking vital signs, performing maintenance and conducting scientific experiments. However, Hockey said that Clarissa, in its early form, handles simple, non-life threatening tasks, such as checking for potable water. But as the system proves itself helpful, the research team envisions adding features.
One of the system's faults had been its inability to discern between voice commands and casual conversation among crew members. So last year, the NASA team brought in Xerox to limit such mistakes.
With expertise in machine-learning techniques and voice-enabled copiers, Xerox was able to give meaning to "fuzzy" documents or sentences that are unclear in translation. It also used text categorization to identify roughly 300 keywords that are command or noncommand words. As a result, the latest version of Clarissa has an error rate of 5 percent, as opposed to last year's 10 percent. However, Hockey said she was unsure when that model would be implemented at the space station.
"It's a lot like spam-filtering, or speech spam, it has to distinguish when you're really talking to it. Xerox technology does a much better job of parsing that," Hockey said.
Experiments with the underlying voice-recognition technology are happening in other fields, too. Manny Rayners of the Ames Research Center is also working with scientists at Geneva University to test the open-source technology in the medical field. The technology helps doctors communicate with patients who don't speak their language.