Imagine the Sounds of Silence

Photo: Charlie Mydlarz, a NYU research scientist, working on an acoustic sensor measuring noise pollution in NYC

SONYC Project Imagines a Quieter New York

It’s possible, if you close your eyes and concentrate, to imagine a New York without exhaust fumes, litter or crime. But without noise? The symphony of sounds—traffic, construction, people, music, bustle, even the guy yelling “hot dog!” at Yankee Stadium—is an integral part of the city’s fabric, but it’s also one of the most prevalent threats to quality of life and health. Most adults in New York are exposed to levels way beyond what is deemed acceptable by the EPA. Noise pollution, when gone unchecked, contributes to stress-related illnesses, high blood pressure, hearing loss, sleep deprivation, and many other challenges to health and wellbeing.

But there’s a fix in play. NYU, in collaboration with Ohio State University, created Sounds of New York City (SONYC), funded by the National Science Foundation. Led by Juan Pablo Bello, associate professor of both music technology at NYU Steinhardt and of computer science and engineering at Tandon, with expertise in digital signal processing, machine listening and music information retrieval, SONYC is a five-year effort to measure sound, collect data, create technology and inform policy to help reduce the noise levels in certain areas of the city.

Up first, the Washington Square area and the region around the Tandon School of Engineering in Downtown Brooklyn, where the researchers are collecting and identifying sounds. In order to get a more diverse survey of sound throughout the city there are currently about 40 sensors attached to NYU buildings in Manhattan and Brooklyn, with plans for more through the cooperation of the Department of Environmental Protection, the Grand Central Partnership, the Downtown Alliance, and other partners.

Eventually, SONYC will have three main components: round-the-clock monitoring with sensors and human input; analyzing that data to extract patterns and behaviors; and the creation of novel data-driven solutions to noise mitigation.

Photo: SONYC researchers test an acoustic sensor

(top of page) Charlie Mydlarz, a senior research scientist at the Music and Audio Research Laboratory (MARL) and Center for Urban Science and Progress (CUSP) at NYU, helped develop the acoustic sensors being used to measure noise pollution in New York City. --- (above) SONYC researchers test an acoustic sensor that is being trained to recognize city noises. From left to right: Juan Pablo Bello, lead principal investigator of SONYC and director of MARL; Justin Salamon, senior research scientist at MARL and CUSP; and PhD student Peter Li.

Simple, Smart Technology
Makes It Possible

The sensors are a little larger than a cell phone. “They look like any other piece of hardware that you may see, like lights or some other electrical equipment, in front of buildings in the city. They have an industrial look,” Bello said.

Each sensor, which is hard-wired into the buildings (they don’t need much power, but they do require a constant stream of electricity), is made up of two working parts. One is a small electronic board that has the microphone, which is capable of converting and pre-processing the incoming audio signal. The other part of the sensor is a tiny, simple Raspberry Pi computer, which computes various sound level metrics and will eventually include the capability to automatically identify sound sources.

The sensors’ two-way communication system piggybacks on existing Wi-Fi systems, either the university’s or available public Wi-Fi networks such as LinkNYC.

The machine-learning aspect of SONYC uses deep convolutional networks, a technology that has been applied to a wide range of problems such as language understanding and visual recognition. “The idea is that these powerful machines can identify time frequency patterns in the sound. You can think of it like a fingerprint. If you have a jackhammer, you have a very specific pattern of rapid impacts over time, which creates a specific fingerprint,” Bello said. “Using these fingerprints, you can train the machines to identify a wide variety of sources.”

Photo of sensor mounted on an NYU building along Broadway in NYC

Acoustic sensors, which are mounted on New York City buildings, are trained to recognize common city noises such as construction.

What’s Next for SONYC

Bello says that in the next year, more sensors will be deployed and more technologies will be integrated into the system to make it smarter. Once there’s an understanding of exactly what sources are creating unnecessary, harmful, or potentially illegal levels of noise, the information will be forwarded to the city for mitigation. The city’s Department of Environmental Protection is tasked with enforcing the noise code in the city, but there are only 50 inspectors to police a city of eight million across more than 300 square miles.

“Obviously, they don't have the scale to be able to enforce all occurrences of noise that are happening in the city all the time. Even if you were to focus on only one noise category, like construction, there are easily more construction sites across the city than there are people to monitor them, and of course you cannot ask inspectors to monitor 24/7.,” Bello said.

That’s where the sensor data comes in—to be able to pinpoint to problem areas and sources. For example, around Washington Square Park, where the initial sensors were installed, the most common noise complaint is after-hours construction. Inspectors could be notified via the sensor network, rather than resident phone calls, and head to the most critical areas first.

“Information from our system can be propagated to those with direct control over noise sources, like construction or building managers, in order to persuade them that noise mitigation is in their, and the city’s, best interest.”

There could also be implications for better processes. Bello gives the example of emergency vehicle sirens—once patterns are detected, the city could divert traffic to mitigate the noise on residents, or to help the vehicles move through faster. “There are many possibilities for how the output of SONYC can be used to control acoustic pollution”.

Photo of acoustic sensor mounted on an NYU building recording construction sounds.

The network of acoustic sensors mounted on New York City buildings will help researchers to monitor, and eventually mitigate, noise pollution in New York City.

The Community Connection

The largest, noisiest and densest city in the United States already has an active citizen reporting system — 311 — so New York is an ideal laboratory to test customized new web and smartphone applications that will tie to machine learning technology,” said Oded Nov, an NYU Tandon School of Engineering associate professor of Technology Management and Innovation who studies human-computer interaction. Nov added, “As we hone the ways to attract and engage citizen participation, not only will New Yorkers help create a quieter environment for themselves and their neighbors, they will also help scientists understand how to involve citizens and technologies effectively in urban environment monitoring projects.” In order for the project to reach its optimal potential, the SONYC team has worked hard to take the added steps of building trust with the community and city agencies. “We are technologists and scientists working to develop what we think are cutting edge, powerful technologies. But in the end, this isn't going to have much value if we cannot interface with the people who are meant to use this technology, or really understand what are the needs of the community,” Bello said. He stressed the importance of citizen participation at multiple levels—contributing data, or even coming in to get training on how the system works.

One of the obvious concerns is privacy. At a time in history when we can assume that nearly any movement we make in public is being filmed by a security camera or tracked via cell phone, should we add eavesdropping to the list? Bello says no. Privacy is an important consideration, baked into the original design and rigorously tested to ensure that conversational content is not captured. Only 10 seconds of audio are recorded at a time at random intervals, never consecutively. Once the technology is fully developed, all audio processing will take place within the sensors, meaning that what is transmitted back to the servers is not a sound clip but a data point noting that a specific sound (such as a siren or car horn) was detected.

The researchers have also taken measures to be transparent with the community by posting yellows signs by each sensor alerting passersby that audio is being recorded. Finally, all audio collected is immediately encrypted before transmission to the project’s secure servers, where it is protected by multiple layers of security.

“SONYC is an environmental pollution platform, is not intended nor designed for surveillance, and we are taking any step necessary to keep it that way,” Bello said.

Bello can also imagine positive byproducts of the SONYC project. “The sensor network will be capturing a significant amount of information pertaining to city life that might not be categorized as noise and that could serve other purposes: for instance, as an indicator of bio-diversity, of usage patterns or occupancy for public spaces informing urban design and policy, to detect traffic accident near misses, as an early-warning system for public disturbances potentially leading to crime, or to assess the effectiveness of mixed zoning.”

A totally silent New York isn’t the goal. “I don't know if I can imagine a quiet New York. I can imagine a quieter New York -- I think it can be better. Noise will always be part of city life, but I think right now our tolerance is too high. We can do better, because noise pollution has a long-term effect on people: on their productivity, sleeping patterns, and the general happiness of the population. I think we can affect all of this for the better if we do a better job of controlling the situation,” Bello said.

To learn more, visit the SONYC website.