Computers don’t always make our lives easier. Consider how time consuming typing can be, or how easily pocket PCs misread handwriting. David Halupka, a U of T master’s student in electrical engineering, is hoping to turn things around.
“I want to get rid of the keyboard,” Halupka says. He is designing a chip that could, along with a pair of microphones, improve the performance level of speech recognition devices.
Less than the size of a stamp, his chip has the capacity to both locate the sound of interest and to separate from background noise.
There are already speech recognition systems that allow you to control your computer by talking to it. “However, they fail when noise corrupts the user’s voice,” Halupka says. With his sound localization and separation chip, he hopes to reduce noise interference, so that speech recognition systems will work well in multiple settings.
Halupka is one of 35 students working at U of T’s Artificial Perception Lab, founded in 2001 by Professor Parham Aarabi in order to research how computers can benefit people.
Equipped with 24 microphones located all around the lab, several web cameras, and numerous workstations, the lab is also home to a robot tour guide. As the robot explains what goes on in each section, the microphones pick up its voice and a computer calculates its location. Once it knows where it is, the robot moves to the nearest station it hasn’t already visited.
The robot looks nothing like a human, though, and Professor Aarabi never wanted it to. There are many researchers who try to make computers look and act like people, but Aarabi doesn’t agree with this approach.
“The human brain is so complicated,” he explains. “Even if we have the same sensors as a human has-eyes and ears, microphones and cameras-we don’t have the computational ability to mimic what the human brain does. If you were to use two cameras and two microphones, humans will always do better.”
Rather than attempting to duplicate man’s mental capabilities, Professor Aarabi chooses to mimic and go beyond human perception. “We try to have sensors that humans may not have, [like] 24 microphones. Humans don’t have 24 ears.”
One interesting application of the equipment at the lab is spatial speech projection. All 24 microphones simultaneously broadcast different sound tracks, and the combination produces different results at different locations in space. A person at one end of the room hears completely different sounds from someone at the other end.
Such processing requires a lot of power. Halupka is interested in low-power designs. He is also interested in convenience, which is why he wants to apply his chip to the palm pilot. “I want to be able to pick up my palm and go, ‘Show me this, show me that,’ just by speaking to it.”
Halupka believes spatial speech recognition could extend to hands-free car controls, smart homes, and anything else that uses voice commands. “It’s time that technology started adapting to us.”
