The Washington PostDemocracy Dies in Darkness

How ShotSpotter locates gunfire, helps police catch shooters and works to ‘denormalize’ gun violence

East Palo Alto’s police chief, Ron Davis, looks at gunshot notifications with engineer Stephan Noetzel. The California city was the first in the U.S. to be completely wired with ShotSpotter. (Mathew Sumner/AP)

ShotSpotter is the audio-sensing technology that alerts police in D.C. and about 90 other cities to when and where a gun has been fired by measuring the distance from its many sound sensors. Within 45 seconds, it can provide not only a detailed location for the origin of the gunfire, but also the actual sounds for an officer to listen to as they speed to the call. Washington Post reporter Clarence Williams sat down with Ralph A. Clark, the president and chief executive of SST, which devised ShotSpotter, to find out the details about how the technology works and what is its future. The interview has been edited for clarity and length.

Can you explain what ShotSpotter is and how it works?

ShotSpotter is an acoustic surveillance technology that detects, locates and alerts on outdoor gunfire, primarily. We do that through a design of sensors that we deploy in coverage areas. These sensors are designed basically to ignore ambient noise, and trigger and time stamp impulsive noises, like booms or bangs.

And so when a gun is fired, that sound — that boom or bang — will emanate, and because we are spacing sensors, our expectation is that three or more sensors will detect that boom or bang, and they will detect it at a slightly different time. And we are able to use that time differential to actually triangulate the location of the original gunshot. And we do all that in a matter of 30 to 45 seconds, so from the time that the trigger is pulled to the time an alert shows up in a patrol officer’s car or a dispatch center — that’s 30 to 45 seconds. We go through a number of classification techniques as well to knock down any potential false positives.

Your technology is essentially a mapping tool, right?

We are leveraging a map, that is for sure. We are resolving to a specific latitude/longitude. So we know the specific location of each of our sensors, and now the other part of the math problem … is to work out the origin of that original boom or bang had to be in this particular location to hit these five or six sensors … at five or six different times.

Once we get that location, which is a latitude/longitude, we can put a dot on a map that has a latitude/longitude, and we can resolve it to an address. The way we get the address information typically is through parcel maps through the agency we deal with. The power of what we do is actually getting officers to a dot, a very specific dot, which could be front of the yard, back of the yard, side of the driveway, etc. The power in that is getting that officer not to just drive by that particular location but to get out of their car. Oftentimes if they are not encountering the perpetrator or aiding the victim, they can recover physical forensic evidence in the form of shell casings, which is critically important with respect to investigating and following up on that shooting.

In those 30 to 45 seconds, can you tell us precisely what is the information that comes to the patrol officer on the street?

Imagine a screen, if you will, and on the right side of the screen is a map. If you are familiar with Google maps or Bing maps, think about a map and think about a dot being shown on a map. That’s one thing we are showing.

Then we will have some metadata descriptions about the particular event on the left side of the screen, such as the latitude/longitude, the address, and the number of rounds fired in that particular incident. Was it a three-round shooting? Was it a six-round shooting? That is critically important. We will sometimes append data based on what we hear when we hear a particular event. If we think there is a multiple shooter event or we think the shooter is moving, our classifiers will append the alert with that additional information, which will show up on the left side of the screen.

And then there will be the wave forms of the specific recordings, and we will have one or two recordings that an officer can listen to. So think about a sound cloud type wave form that you see, you can press the button, and the officer can actually hear a recording snip of a particular event so they know exactly what they are going into. Actually having an officer hear that sound can be extremely powerful and motivating to get to a particular location very, very quickly.

From an officer safety point of view, it’s a game changer — to go from 89 percent of gunfire sadly going unreported by the communities that suffer the most, to a police department having complete awareness any time a gun is fired. To be able to have it delivered very quickly and very precisely, in terms of a very specific location, with the actual full context of the event so they know what they are stepping into.

What are some of the challenges that lead to inconsistencies in terms of real time reporting?

We strive to do a really good job in not having false negatives, i.e. missing gunshots, and not producing false positives. On the latter aspect, we kind of go through a two-step classification process. There is a machine learning algorithms that can distinguish, for the most part, gunshots vs. non-gunshots, and then we also have a separate human review process where humans will actually listen to and look at these events and make the final determination if it’s a gunshot or not. That’s what really makes it a 30- to 45-second trigger pull to alert showing up vs. a 15- to 30-second notification process. We’re spending an extra 10 to 15 seconds having human reviewers, who we train to make those determinations.

The one thing I would say is that it took us a while to convince them [D.C. Police] that we could do a better job of classification than they could. Although they were very good, just by the nature of it, they couldn’t be as good as we are because that’s all we do. Our folks aren’t distracted with anything else. We train them to do that. They get to see gunfire over 90 cities, over a long period of time. So they really, really are expert at it.

In our relationships with our customers, we try to drive a high degree of intimacy. Which means when we are helping out, we like to hear positive and constructive feedback. But we’re equally passionate about hearing where we have produced a false positive or even a false negative. That information is very helpful to us to kind of go back and [use] remedial efforts and add sensors. It’s not a perfect system, that’s to be sure. We have a very high Service Level Agreement standards, but we know invariably in the world of trying to detect, locate and precisely alert on gunfire, we are going to have some misses and some misclassifications. That’s the nature of the beast. Our SLA standard is 80 percent. That is too low for me. If we are at 80 percent, we really suck frankly.

Can you tell me anything specifically about the District’s contract or how ShotSpotter works in the District?

I always defer to our agency partners to talk about how they use the system specifically. I guess I would make the more general commentary that the thing we strive most around our relationship with customers is to think about how they respond to and investigate gun violence in a very different way.

If you think about what we do, we digitize and automate the notification, location and alerting of gunfire, which is a very analog, messy process right now. It doesn’t happen 80 to 90 percent of the time because it’s become so normalized in communities, where people don’t call [911], and when they do call, it takes a long time for them to get through. They don’t have a precise location to send anybody to. They are certainly not having a digital recording of the event. In this digitized world, … you can almost think of our system as being a robot if you will. It is always detecting, precisely locating, with very rich metadata descriptions of the number of rounds counted, digitized recording, etc. And this is happening in real time, 30 to 45 seconds, that’s game change right?

Now that we have this ability, how can you respond differently to gun violence? For me it’s like equalizing the service level agreements that police departments have with their communities. We all know if a gun was fired in Georgetown, you’re going to get a very robust response. Anacostia deserves the same response that Georgetown does. Even if the residents don’t call 911.

Officers need to get there, get out of their car, and if they are not encountering a perpetrator or aiding a victim, after they recover the physical forensic evidence, they are knocking on doors, asking, ‘Mr. Clarence, are you okay? We know there was some gunfire here, we’re really concerned about you.’

When you do that, that changes the relationship between the community and the police. You are showing up in service.

A lot of times people want to look at our technology and say how many arrests do you make because of a ShotSpotter activation? They don’t really get what this thing is about. The bigger power is in not arresting a criminal when they are firing the gun, but preventing the criminal from firing the gun in the first place. The only way you do that is to begin to denormalize gun violence and get the community to buy into the fact that gun violence is unacceptable and shouldn’t happen. That’s a very philosophical thing, but it’s very, very powerful.

Getting agencies and cities to buy into that is really what we have to be about. Not all of them embrace it fully. The first thing is that you have to acknowledge the pain that these communities are feeling. That’s what our system does unabashedly. We don’t need any help from anybody to transparently report on the incidents. Once those incidents become more transparent you have an obligation, a responsibility to do something about that. And sometimes that’s a bit of a delicate dance … with the police department, generally speaking.

What do you consider to be success stories and can you describe them?

This is much more of a preventive exercise rather than catching criminals with guns in their hands. We do an annual gunfire index, we’ve seen 35 percent reductions on year-over-year gun violence, I think that’s a very strong indicator of the impact.

In an individual city?

Typically, after two or three years of operations, we are seeing those kinds of very large double-digit reductions in gun violence in the coverage areas, where we’re deploying. We are very careful and humble not to believe we are the complete straw that stirs the drink with respect to that. In a very self-selection process, we are part of an overall, comprehensive strategy. The reason we are going to be successful is any chief and city that adopts our technology probably has a more forward leaning posture on gun violence reduction. So they’re not just doing ShotSpotter, they’re doing other things as well, and all those things together are contributing to these declines.

We leverage mapping … We know our coverage area, we collaborate with the agencies who say I’m interested in covering five square miles. We like it to be as close to a circle as possible to make it more efficient for our sensors. We are deploying 15 or 20 sensors per square mile. We are responsible for maintaining the sensors because we are providing a service, we aren’t asking the city or agency to buy equipment and maintain sensors. We are responsible for getting the permissions to deploy our sensors on rooftops and the like. That’s what it’s about.

Have you been a part of successful prosecutions or a part of the prosecutorial process?

That’s a really important part of what we do. We have a full-on detailed forensic report, with all the chain of custody and all that stuff. We have expert witness resources available to go and testify in court. We have three people in our company that all they do pretty much is fly around the country and support district attorneys, who weren’t successful in getting DA advantaged plea bargains. For those cases that do go to court, we are participating in a number of those and have survived all challenges to that. It is forensically accepted in over 17 states and federal courts …

The digital world is very precise, very fast. You can’t argue the facts and stuff. It’s not your interpretation vs. my interpretation. I can say, two rounds fired here at this time. I can let you listen to it. I can send it around to other places if we need to and have other people collaborate around that. It’s extremely powerful …

We are raising soldiers in some of these communities. That never gets captured in any UCR FBI data. It just doesn’t. When people talk about capturing criminals with guns, we’re measuring the wrong thing, I think. The thing that you’re not measuring that has much more impact, which is parents, families raising kids to be soldiers. What’s the output of that? Dropout rates, teenage pregnancy, alcohol abuse, drug abuse, violent crime, etc. I have been raised to be a soldier. My expectations and options have been cut off to me.

That’s the thing we want to impact. We don’t want families raising soldiers. They shouldn’t have to.