You Can Help Train AI to Pinpoint NYC's Noisiest Neighborhoods in an Effort to Shut Them Up

New York City is very loud, as you might infer from a place whose principal mythology insists it never sleeps. Or you might know firsthand from having lived there and somehow never managed to stray more than three blocks from a clanging jackhammer or an incessant car alarm. It’s so loud that in 2014, a city councilwoman authored a bill designed to flag areas that consistently saw noise levels above 65 decibels; about the level you’d have to raise your voice to carry on a typical conversation—and enough to pose an actual health issue to those continually subjected to the din.

To try to better understand the precise nature of all this noise, researchers with NYU’s Sound of New York City project are hoping to employ machine learning algorithms taught by citizen scientists. SONYC has spent the last two years recording audio data around the city, especially near construction sites and other high-volume areas. According to the university, they’ve collected about 30 years worth of literal sounds of New York.

Now, SONYC is asking citizens with a stake in rendering New York relatively quieter to log onto its new website and identify snippets of that sound, to help train a machine learning algorithm to understand what it’s listening to. Sifting through all the material would be “an impossible task” for researchers, Mark Cartwright, a postdoctoral researcher at NYU’s Music and Audio Laboratory, tells me.

“There simply aren’t enough resources available to have people manually coding all this data,” he says. “It would require hundreds of people sitting at computer screens 24 hours a day, seven days a week. Machine listening models will help to automatically categorize reams of sound data, which will be crucial for us to start conducting large scale analyses on the sound data and determine patterns of noise over time.”

This, in turn, will help the team create “aural maps” of New York, which can be used to understand what, exactly, is responsible for making neighborhoods noisy. Sometimes it will be obvious—construction sites—sometimes, less so. SONYC notes that surveys of 311 calls show noise complaints to be a major quality of life concern and that 70 million Americans live in places that are exposed to higher noise levels than considered healthy by the EPA. They point to sleep disruption, hypertension, heart disease, and hearing loss as primary concerns in New York. “It’s clear that noise mitigation is in the public interest,” Cartwright said.

Users will listen to a sample and select the best description of what they’re listening to. When I logged on, I heard what would definitely elicit a “siren” click. As of writing, nearly 800 volunteers have classified almost 13,000 noise samples, which seems like a fine start. New York residents and anyone else who’s interested can log in here and start listening in on random street corners of New York. (Researchers say the data has all been anonymized.)

“This knowledge will help us to identify noise patterns and its causes over time, which can be used by New York City residents and policymakers to improve noise pollution in the city,” Cartwright says. “If we can send automated feedback on sound sources and levels to say, the manager of a construction site, we can empower those people to take action to reduce their noise footprint and adhere to the noise code.” They hope to share the data with city noise inspectors, so they might better locate violations and when they are occurring. “Existing technologies are unable to isolate the effect of offending sources,” Cartwright tells me.

Eventually, he hopes the model will “help improve our future cities by informing urban planning around schools, traffic routes, parks, and more.”

Some have voiced concerns that machine learning will stamp out citizen science, but this seems to be a unique bridging of the two and a concrete example of AI deployed for the public benefit. I considered ending this with an aural AI pun—we involved in the machine learning space should be all ears for the results—but quickly thought better of it. Instead, I’ll just say that with a little help from machine learning, the city that never sleeps may finally be able to enjoy the sounds of silence.