Coast Guard using voice tech to help catch hoax caller

"There's infinite amounts of information that's buried in the speech"

A Coast Guard helicopter is sent out for a rescue, file. (Cannon Beach Fire & Rescue)

PORTLAND, Ore. (KOIN) — The Coast Guard takes calls to their local stations very seriously. Each time a call comes in, a giant, expensive search and rescue operation is kicked into high gear. But lately, more and more of those calls are turning out to be bogus.

Lt. Chris Bodner said he’s serious about hunting down the person who made 22 false mayday-type calls to the Coast Guard’s Warrenton station just in the last year.

A telephone keypad (Public Domain Photos, sevenheads lopez)
A telephone keypad (Public Domain Photos, sevenheads lopez)

“It becomes an issue when the intent is to deceive the Coast Guard into thinking there’s distress when there is not,” Bodner told KOIN 6 News.

Officials believe the calls are coming from Chinook, Washington or somewhere between Warrenton and Seaside.

To help them catch whoever is responsible, the Coast Guard is turning to a new kind of technology that uses voice recordings to create a physical profile of the caller.

“It’s a powerful tool, it’s a valuable tool,” Dr. Rita Singh said. “You can get a personality, you can get demographic parameters, race, origin of the speaker.”

Singh, a computer scientist at Carnegie Mellon University, is the lead developer of the technology. It creates a physical profile from analyzing a single voice and through the deduction of physical parameters like height, weight, body shape and facial structure.

Listen: False Coast Guard mayday call

“It’s also the deduction of physiological parameters, age and the presence or absence of medications in the body,” Singh said. “Behavioral parameters, whether this person is a leader. What are the dominant characteristics, medical characteristics?”

Singh’s computers are loaded with cutting-edge, machine-learning algorithms and signal process technologies that use a bank of predictors to arrive at a profile.

Dr. Rita Singh developed technology that uses voice recordings to create a physical profile of the caller. (Carnegie Mellon University)

“It’s like a bunch of experts making predictions about something, and at the end of the day we have to apply more algorithms to come up with a final prediction based on their predictions,” she explained.

Singh analyzed clips from the hoax calls to Warrenton, knowing only that they came from the Coast Guard, but not specifically where. She concluded the culprit is a white man between 35-40 years old, 5’10”-6′ tall and weighing about 190 pounds.

“Even this much information is valuable information for law enforcement agencies,” she said. “They’re looking for any and all information they can get about the criminal.”

Still, not everyone is completely sold on using voice analysis for physical profiling.

“Speech, in many ways, is a very messy signal. There’s infinite amounts of information that’s buried in the speech,” Jan Van Santen, director of Oregon Health & Science University’s Center for Spoken Language and Understanding, said. “That is a growth business, but I think the jury is still out about whether that really works.”

He said there’s interest in using voice for legal purposes, like fingerprints and DNA.

While Singh admits the technology is still in its infancy, she hopes it will vastly improve in the next 3-5 years. Until then, the Coast Guard believes it’s at least worth trying to use it to crack the Warrenton hoax caller case, whether it works or not.