Speech Technology Magazine

 

Out of Harm's Way

In-car speech system adds a layer of safety for police on the beat.
By Ryan Joe - Posted Jan 25, 2008
Page1 of 1
Bookmark and Share

The most common fatalities for police officers in the United States are the result of gunfire and automobile accidents. As of press time, 64 officers had been fatally shot in 2007, according to FBI statistics, while fatal automobile accidents have seen a 15 percent increase over 2006 figures.

These statistics underscore the dangers that police officers face daily. To improve officer safety, the University of New Hampshire (UNH) and the New Hampshire State Police, with funding from the U.S. Department of Justice, have been collaborating on Project 54, a state-sponsored program that outfits police cars with a speaker-independent voice interface to control many in-car systems.

Project 54, so named after the TV show Car 54, Where Are You?, started in 1999. It uses Microsoft’s SAPI 5 to speechify patrol cars, giving officers eyes-off, hands-off operation.

Using the system, an officer can pull over a vehicle, run the license plate, and, seconds later, receive an entire record of both vehicle and driver. That’s because Project 54 equipment is linked wirelessly to the databases of dozens of local, state, and federal enforcement agencies. 

Previously, the same officer would have to pull the car over, radio in the license plate, and wait in the queue with all the other calls coming into the precinct or command post before receiving the information. "It would take anywhere from five to 20 minutes to get a response," says William Lenharth, director of UNH’s Research Computing Center and Project 54’s co-director. "And they’d just be sitting in a car with the guy in front of them arming himself."

But aside from these rare and sudden acts of violence, even the simple act of driving can be dangerous for police. Common distractions are compounded, particularly during evening and overnight tours when officers take their eyes off the road to scan alleys or parking lots for questionable activity.

Additionally, patrol cars have a litany of in-car equipment, including lightbars, sirens, video cameras, cell phones, pagers, radar equipment, GPS units, and two-way radios with as many as 512 frequencies, all of which would otherwise need to be controlled manually.

With so many radio frequencies alone to monitor, it "becomes somewhat cumbersome trying to find a particular channel," says New Hampshire State Police Lt. Mark Liebl, who field-tested the first Project 54 car in 2002. "Particularly if you’re responding to a critical incident—shots fired, officer down, robbery in progress—and you’re going at a high rate of speed with lights and siren, cognizant of traffic, some people don’t pull over as quickly as you’d like. Suddenly, you get instructions from the communications center to go to SWAT3, go to TAC2, go to the City of Portsmouth’s or the City of Concord’s frequency for further instructions."

With Project 54, an officer needs only to  hit the push-to-talk button on the steering wheel, say the command "Coast Guard," and the system automatically locates the frequency. The same button also unites the controls for lightbars, strobes, sirens, radar, and many other devices under a single voice interface.  

To keep system costs down, the UNH faculty and grad students who worked on Project 54 used off-the-shelf components. "If a car already has a computer and a lightbar that’s compatible, you would spend about $1,200 installing it, and you could take it from car to car," Lenharth says. "You could hold all the parts in your hand."

The software needed to run the system costs $500 per department, regardless of the number of vehicles in its fleet. Because of a state grant, it’s free to departments throughout New Hampshire.

Growing Interest
Lenharth estimates that there are roughly 1,000 Project 54-equipped vehicles on the streets nationally, about 885 of them scattered among 190 departments in New Hampshire alone. "There’s probably 60 cars in several departments in California," he says. "The biggest is in Carlsbad; they have 55 cars. San Diego Fire Department has a vehicle we just did. Maryland State Police has over 80 cars. Fairfax County [Va.] Police have around five cars. Burlington, Mass., Police have maybe eight or nine."

Project 54 has the potential to go beyond patrol cars. The system is being tested in a number of patrol boats, ATVs, fire and other emergency vehicles, and even motorcycles.

Liebl, who has taken his Project 54 car to showrooms at police conventions nationwide, has seen interest from departments in places as diverse as Kentucky, Germany, and Australia. "Having used it for as many years as I’ve used it, when I do lose my system or find myself in a cruiser that hasn’t been outfitted, I miss it," he says. "I truly miss it."

Developing the Project 54 system was no easy task at first. Perhaps one of the biggest hurdles for UNH teams was the proprietary nature of many of the in-car systems in use.  Lightbars, radar, and radios, for instance, can come from several different manufacturers; project leaders had to convince the myriad vendors to unveil the source code for their respective products.

Another problem that developers had to overcome was environmental—highway driving versus driving through an inner-city neighborhood, for example, and background noise that can affect the performance of any speech system. It’s something that UNH researchers are still working to perfect.

Currently, the Project 54 speech interface maintains an 85 percent task completion rate. In looking at the problems, researchers found that one-third of recognition errors were attributed to speech recognizer problems. The other two-thirds were attributed to user error, such as not speaking loudly enough, issuing an invalid command, issuing a valid command out of context, or operating the push-to-talk button incorrectly.

In experiments conducted using Project 54-equipped cars, researchers found a direct correlation between recognition rates and driving performance. "We tested extreme cases,—44 percent and 88 percent recognition," explains Andrew Kun, an associate professor in UNH’s Electrical and Computer Engineering Department. "And we found that the lower recognition, compared to higher recognition, does have a negative effect on your driving performance. And if you have a low recognition rate, then pushing the press-to-talk button reduces your driving performance."  

Another Project 54-related test involves the use of a graphic user interface (GUI) on a GPS unit. UNH researchers are particularly interested in how looking down at the screen and manually entering information influences driving. They also want to find scenarios in which drivers are likely to scan a GUI and the sort of feedback that will reduce the likelihood of them doing so.

Liebl is currently testing a voice-enabled GPS unit in his patrol car. "I don’t have to type anything in," he says. "I don’t have to look away from the road or take my hands from the wheel."

It seems like a novelty at this point, but Liebl is cognizant of the much greater stakes. "It’s not about what’s in it for us," he says. "It’s about what’s in it for law enforcement. And in the bigger picture, what’s in it for public safety."

Page1 of 1