A personal trainer in the palm of your hand
You get a call on your cell phone. Your personal trainer wants a report on your physical activity today and what you have been eating.
In short order, you get feedback on how the exercise you've logged fits with your goals for the day, and on how much you have left to do, with suggestions for how to meet your goals. You could even get an assessment of the nutritional value of your diet.
You also might receive a reading on whether the activity was good for your bone density, your brain and other parts of your body.
"I think there's so much possibility," University of Illinois Professor Weimo Zhu said recently. "It's like hiring a personal trainer. But this will be much cheaper, easier to do. It really could become like an e-trainer, I call it."
The "trainer" is a computerized system developed by Zhu, a UI kinesiology and community health professor, and colleague Mark Hasegawa-Johnson, an electrical and computer engineering professor.
At the center of it is software that converts sound bites users record about their physical activity to exercise data, figures the amount of energy and calories expended, compares it to desired levels and otherwise turns the recordings into useful information automatically.
"The goal is to make it as easy for the user as possible," said Hasegawa-Johnson, whose research focuses on speech-recognition technology.
Zhu and Hasegawa-Johnson – along with computer science Professor Dan Roth, a text processing expert, and graduate students Mital Gandhi, Art Kantor, Yong Gao and Youngsik Park – already have tested the system on more than 30 volunteers.
The participants carried a standard digital voice recorder and were prompted to log their activity periodically throughout the day by an alarm watch they wore.
The digital recordings were then uploaded to a computer and analyzed by a package of commercial speech-to-text software and the activity-categorizing and -analysis software developed by the UI researchers.
"It actually seems to be working pretty well," Hasegawa-Johnson said. "The biggest problem is background noise. People are using these in all kinds of environments."
Zhu said the system isn't yet quite as good as humans at categorizing the activity reports, but it's close.
Next, the UI researchers plan to test the idea of using the system with cell phones rather than digital recorders and alarm watches. They've received a grant to develop a prototype automated call center, which should be ready this fall, Zhu said.
The idea is that the call center will ring users at intervals, collect a report on their activity, process the information and, eventually, be able to provide instant feedback. Both Zhu and Hasegawa-Johnson think the system has commercial possibilities.
It came about, however, because Zhu wanted a better way to collect data for his research. He's an expert in "kinesmetrics," a field which applies measurement theory, statistics and mathematical analysis to the study of physical activity and health.
His studies often relied on participants keeping paper and pen activity diaries, which then had to be entered into a computer. The method was labor intensive and open to data-entry and categorization errors. Not to mention that many participants just got sick of keeping the diary after awhile and quit.
"You try to decrease the subject's burden as much as possible," Zhu said. "Otherwise no one will use it. That's the key, key thing."
He hit on the idea of a system using voice recognition with automatic data classification and went looking for collaborators at the UI. The researchers have been working on the system since 2003, with funding from the Robert Wood Johnson Foundation and the Neer Research Fund at the UI.