A teen uses math to pick out a voice in the crowd
WASHINGTON, D.C. — When his grandfather lost his hearing two and a half years ago, George Hou, 17, thought hearing aids might help. But his grandfather hated the devices and refused to wear them. His hearing aids didn’t increase the volume of only the sounds he wanted to hear. They upped the volume of everything, surrounding him with noise. George, a senior at Arcadia High School in California, thought there must be a better way. He worked to invent a mathematical model that can picks voices out from a background of noise.
“He’s a cool guy, we’re really close,” says George of his grandfather. But once his grandfather stopped wearing his hearing aids, “We could only communicate via pen and paper,” the teen says. “I was really devastated.”
George decided to find out more about how hearing aids work. He hoped to figure out how to help listeners hear speech without drowning in background noise.
His first step: He picked up a book called Mathematical Modeling and Signal Processing in Speech and Hearing Sciences. If that title sounds very complicated, that’s because it is. “I flipped through it,” George recalls. “But the only things I understood were the abstract [the summary] and the conclusion.”
So the teen emailed the book’s author, Jack Xin. He’s a mathematician at the University of California, Irvine. Xin wrote back, giving George some papers to read. Soon, the two started working together to develop a mathematical model to could pick sounds out of a noisy crowd.
There are many ways that computers can try to separate a single voice signal from background sound. Many people have created computer models —simulations — that filter out background noise or dampen it. But as they filter out noise, the devices also can filter out things that people want to hear.
Some programs filter out noise using a method called spectral subtraction. It compares the spectrum — or range of sounds — within a sound signal that the program is looking for (voice) to the range of all other sounds (noise). By subtracting the spectrum of the background noise, only the signal should remain.
For example, programs can filter out sounds that have a high frequency. To the human ear, these will have a higher pitch. But sometimes these sounds are not ones someone would want to miss — such as the high-pitched voice of a child crying or the wail of an ambulance.
George and his mentor came up with a model that can take a noisy scenario and separate the sounds into a signal and the background noise. Instead of trying to filter background noise out, the model treats it the same way as the voice that the model is meant to detect. By doing so, the model allows that signal to be separated out more effectively.
Xin and the teen put their model into a computer program. Then they used it to pick clear sounds out of a noisy sample. George published his research. It appeared January 15, 2015 in Society for Industrial and Applied Mathematics Undergraduate Research Online.
George’s method is still just a model on a computer screen. It has not been tested even in a real situation, such as a cocktail party or coffee shop. It also needs to be improved so that it could find and suppress signals in real time, not just in a recording.
The teen’s work, however, took him all the way to the Intel Science Talent Search. This program is run by Society for Science & the Public and sponsored by Intel Corp. Each year, 40 high school seniors are chosen to come to Washington, D.C., where they share their science with the public and compete for cash awards that currently total more than $1 million.
Learning about how to separate sounds gave George an unexpected respect for math. “Growing up, I didn’t like math at all,” he explains. “It wasn’t until my research that I began appreciating math. I wouldn’t say I love it,” he says. But finally, he adds, “I appreciate what it can do.”
Follow Eureka! Lab on Twitter
(for more about Power Words, click here)
abstract Something that exists as an idea or thought but not concrete or tangible (touchable) in the real world. Beauty, love and memory are abstractions; cars, trees and water are concrete and tangible. (in publishing) A short summary of a scientific paper, a poster or a scientist’s talk. Abstracts are useful to determine whether delving into the details of the whole scientific paper will yield the information you seek.
computer model A program that runs on a computer that creates a model, or simulation, of a real-world feature, phenomenon or event.
frequency The number of times a specified periodic phenomenon occurs within a specified time interval. (In physics) The number of wavelengths that occurs over a particular interval of time.
hearing aid A device designed to amplify sound for a wearer. They are usually designed to make it easier for the wearer to understand speech.
Intel Science Talent Search An annual competition created and run by Society for Science & the Public and sponsored by Intel Corp. Begun in 1950, this event brings 40 research-oriented high school seniors to Washington, D.C. to showcase their research to the public and to compete for awards.
model A simulation of a real-world event (usually using a computer) that has been developed to predict one or more likely outcomes.
pitch (in acoustics) The word musicians use for sound frequency. It describes how high or low a sound is, which will be determined by the vibrations that created that sound.
simulate To deceive in some way by imitating the form or function of something. A simulated dietary fat, for instance, may deceive the mouth that it has tasted a real fat because it has the same feel on the tongue — without having any calories. A simulated sense of touch may fool the brain into thinking a finger has touched something even though a hand may no longer exists and has been replaced by a synthetic limb. (in computing) To try and imitate the conditions, functions or appearance of something. Computer programs that do this are referred to as simulations.
Society for Science and the Public(or SSP) A nonprofit organization created in 1921 and based in Washington, D.C. Since its founding, SSP has been not only promoting public engagement in scientific research but also the public understanding of science. It created and continues to run three renowned science competitions: The Intel Science Talent Search (begun in 1942), the Intel International Science and Engineering Fair (initially launched in 1950) and Broadcom MASTERS (created in 2010). SSP also publishes award-winning journalism: in Science News (launched in 1922) and Science News for Students (created in 2003). Those magazines also host a series of blogs (including Eureka! Lab).
spectral subtraction This is a mathematical method of separating out a signal from a noisy background. The method uses a sample of sound, and compares the range — or spectrum — of a sound in the intended signal to the range of sounds in the background noise. By subtracting the range of the noise from the range of the signal, only the signal should remain.
spectrum (plural: spectra) A range of related things that appear in some order. (in light and energy) The range of electromagnetic radiation types; they span from gamma rays to X rays, ultraviolet light, visible light, infrared energy, microwaves and radio waves.