Ah, mais oui. (But the data would probably also stand up in court.)
A few weeks later, back in the States, I visited the headquarters of a smaller but similar noise-monitoring project, at N.Y.U.’s Center for Urban Science and Progress, on Jay Street, in Brooklyn. That project is called Sounds of New York City (sonyc) and is funded mainly by the National Science Foundation. sonyc’s purpose, Mark Cartwright, one of the scientists on the project, told me, is “to monitor, analyze, and mitigate noise pollution.” Each sensor in its network has just one microphone, which is roughly eight inches long and covered in foam. The microphone is attached to a small, weatherproof aluminum box, which also contains a Raspberry Pi. Sometimes the sensors are mounted with a long strip of plastic spikes, which are meant to deter pigeons from using the devices as latrines, and which, on monitors installed near Washington Square Park, have developed the unanticipated additional function of accumulating tangled masses of the wind-borne hair of N.Y.U. students.
The method that sonyc uses to collect data and to document noise-code violations is different from the one used by Bruitparif. The sonyc researchers are developing algorithms that they hope will eventually be able to identify a full range of noise sources by themselves—an example of so-called machine listening. “Having a network of sensors deployed around the city enables us to start understanding the patterns of noise and how they develop around things like construction sites,” Charlie Mydlarz, another scientist on the project, told me. He said that sonyc also gives the city’s Department of Environmental Protection actionable evidence of violations. Mydlarz and his colleagues are still training their algorithm, with help from “citizen scientists,” who visit a Web page and annotate ten-second audio files, collected by the sensors, with what they think are the sounds’ likeliest sources: jackhammer, car alarm, chainsaw, engine of uncertain size. He demonstrated the algorithm’s current iteration by alternately operating a Black & Decker electric drill and the siren of a toy fire truck near a sensor on the table in front of him. The algorithm successfully identified each and measured its decibel level. (It can also identify the fire truck’s horn.)