Listening for Gunshots
An alert pops up on one of Liz’s six computer screens. She clicks on an image of the sound that triggered it — a green staccato waveform generated by something happening right now, something loud and implosive and nearly 3,000 miles away — and listens intently.
Liz squints. “Trinh, that’s hammering, yeah?” she says.
Another analyst leaves her own array of screens to listen as Liz plays back several snippets of audio from different sensors near the source of the bangs. They agree that it doesn’t sound like gunfire, but like someone banging a hammer — on this August afternoon in Paterson, New Jersey, perhaps someone is fixing a roof. Liz dismisses the alert.
The next alert is for a stranger sound, and Liz rejects it immediately. “It’s raining in Bayamón,” she says of the town in Puerto Rico. The droplets are hitting the sensors: bang, bang, bang. The next is from Canton, Ohio, and once again the bang — this one muffled and accompanied by the mournful sound of a train whistle somewhere in the distance — doesn’t sound like a gunshot, but maybe something large and metallic being dropped.
Liz is nearing the end of her eight-hour shift at ShotSpotter, a company that sells its audio surveillance systems, which record and triangulate the sounds of gunfire in high-crime areas, to police departments around the country (and a few abroad) that want to know precisely when and where guns are going off. Its microphones cover more than 300 square miles in more than 90 cities, most of them linked by the web to this small room in an anodyne office park near the southern tip of San Francisco Bay. The technology is designed to be able to tell the difference between gunshots and other similar sounds, but the company also has humans on duty 24 hours a day, determining which bangs are actually gunshots that police on the ground should know about.
The training lasts six weeks and involves listening to hundreds of recordings of fireworks, helicopters, truck brakes, gunshots: all the clamorous din of modern cities, extreme decibels stripped of their context. Some are just trashcans banging together, but some are the sounds of death.
“It’s not, you know, a normal job,” says Ginger, another analyst, who used to work as a teacher. Most of the time, she says, “I’m just hearing noises and trying to figure out what it is” — focusing on the task at hand rather than thinking about what happened or why or whether anyone was hit, about what might be unfolding in the wake of the shots she heard. “I guess,” she says, “if I sit and think about it, it’s pretty horrifying.” Two days ago, Ginger alerted the Oakland police, just up the bay, that 15 rounds had been fired in a neighborhood called Hegenberger. She heard later on the police scanner that a woman had been shot in the head and had not survived.
On the wall overhead, a large screen shows the room’s latest statistics: In the past week, analysts have screened 14,415 possible shootings and reported more than 1,500 of them to police as gunfire. In the past month, they’ve reported more than 6,000 shootings. By the end of the year, the company’s CEO, Ralph Clark, thinks they’ll have detected 100,000 gunshots in their coverage areas alone.
If that sounds like a lot of shooting, it is — far more gunfire than we once thought was happening. The year after Miami installed ShotSpotter systems in three neighborhoods, it saw reported gunfire shoot up 800 percent — not because the neighborhoods had suddenly become more violent, but because the surveillance system was picking up shots that weren’t otherwise being reported to the police. Across the country, ShotSpotter’s microphones reveal gunfire to be a much bigger component of city sounds than officials had guessed: When the company compares its data to 911 calls, it finds that, on average, only about 20 percent of gunshots are reported.
Clark, a black man who grew up in crime-ridden East Oakland but now lives in an affluent area of the city, says this divide is indicative of “a tale of two cities”: of neighborhoods where one shot will spawn ten anxious 911 calls, and others where gun violence has been so normalized, and police have earned so much distrust, that no one will call. In such neighborhoods, says Clark, “there’s cumulatively a mass shooting that happens every day in this country.” But that kind of violence tends to be more invisible than the kind that startles places where gunshots are rarely heard.
He shows me his phone, where he has a map updated with dozens of shootings, flagged by the number of shots fired per incident, in ShotSpotter zones in nearby San Francisco and Oakland during the past week. It’s a strange window into the usually hidden geography of violence; I realize I walked right past the site of a recent shooting on my way to get breakfast a few hours earlier.
ShotSpotter has mostly stayed out of the politics of guns and gun control but has commonly been on the receiving end of other criticisms, including about its efficacy. In San Francisco, for example, the Center for Investigative Reporting found that 3,000 alerts of shots fired led to only one gun-related arrest; most people aren’t dumb enough to hang around after firing a gun. Clark says this kind of framing misses the point, that the system isn’t designed to lead directly to arrests but to help police departments better understand what’s going on in their jurisdictions — where, why, and when gun violence is happening, plus who has and shoots guns. (He repeated a line I’d already heard, verbatim, from a ShotSpotter employee: “This isn’t CSI: Miami where every case gets solved in 30 minutes.”) Clark also thinks having ShotSpotter in place lets cities denormalize gun violence, with police showing up to investigate whenever a gun goes off, and increase trust by indicating to residents of underserved neighborhoods that the police do care about what goes on there. Still, a few police departments have canceled their contracts, saying that the promises of big data are oversold and that their money would be better spent on old-fashioned community policing.
Critics also argue that the company’s sprawling microphone systems are a troubling invasion of privacy. In recent years, the company has denied police requests to hear audio captured before or after the short snippets of gunfire that it shares. (“That seems really friggin’ spooky,” says Clark, who put a sensor on his own house in an effort to tamp down the criticism. “We want not to be Big Brother.”) The company muffles its sensors so they pick up more gunfire and less ambient sound, and deletes everything but recordings of actual bangs within 72 hours (to check why gunshots that were reported in other ways, but missed by ShotSpotter, didn’t trigger the system), a policy it developed after discussions with the ACLU. In 2015, Jay Stanley, a senior policy analyst with the ACLU, summed up his thoughts on the privacy implications of the system: “I am not losing sleep over this technology at this time,” he wrote. “But I am concerned over the precedent of allowing our cities to be sprinkled with live microphones that are not subject to transparent operation, and where that will lead over coming years and decades.”
Back in the Incident Review Center, where it’s still the morning shift, the slowest time of day, Liz dismisses more alerts, including what sounds like a garbage truck slamming down a can in Detroit, and something she can’t identify but doesn’t think is a gunshot in Cape Town, South Africa. There, the company’s microphones are placed lower than in most cities, so the short clip is strangely intimate: The sensor picks up a motorcycle going past, what sounds like a kid laughing, a barking dog.
The analysts at ShotSpotter are used to seeing, and trying to make sense of, a big world through these tiny snippets of sound. They can point out a certain small, forested section of San Juan, Puerto Rico, that they think is a practice range for automatic weapons, and wonder why people in Denver seem to like fireworks so much. They pay close attention to shifts in weather and seasons, which they see reflected in gunfire: People stay inside, and shoot less, when it’s too cold or too hot or too wet. Summer weekend nights are the busiest times of the year, unless you count the Fourth of July and New Year’s Eve, when there are so many fireworks to sift through that the company presses managers into service and sets up listening stations throughout the office.
They come to know, and allow for, the different quality of sounds in the places they monitor: the echoing canyons of buildings in New York City, the long distances sound can travel in Omaha, the way humidity and water muffle noises all along the East Coast. They learn to listen for the “boomy, hollow” sound that distinguishes fireworks from the cracks of gunfire, or the buzz of a saw that might indicate the bangs of a construction site. But the sounds usually stay abstract. Liz tells me that she takes a longer than necessary route into work each day to avoid one of the company’s coverage areas in East Palo Alto; the fact that the company’s sensors are usually set up in high-crime areas means she makes an effort never to visit in person the places that she monitors.
When an alert for actual gunfire comes across Liz’s computer, she knows immediately that it’s real. “That was Oakland,” she says, consulting the sensors’ triangulation on the map on her screen. “Three shots in a backyard.” We listen to the shots ring out but can only guess what happened, whether anyone is hurt. The next day I’ll find a short news item saying that someone was taken into custody for firing a gun in the air.
Liz’s shift ends, and the afternoon shift begins, a new group of analysts taking their seats in front of the arrays of screens. Ginger plays me the recording of the shooting she alerted the Oakland police about two days ago, the 15 rounds that led to a woman’s death. We listen to the sound — all those haunting, relentless bangs — of what we know happened. It sounds just like, and also nothing like, the many gunfire samples that analysts have played for me today: the soundscape of American cities in anonymous bangs, dislocated from whatever came after.