Pardon the Interruption

Apartment living has its ups and downs, with the downs often winning out. Living in close proximity to perhaps dozens of other people, each with different schedules, tastes, and ideas about what constitutes acceptable behavior almost inevitably leads to clashes and compromises. Noise, in particular, becomes a constant issue, whether it is the neighbor's late-night parties, the incessant barking of a nearby dog, or the cacophony of footsteps echoing through thin walls. Privacy feels like a luxury as every sound, from conversations to the opening and closing of doors, infiltrates one's living space.

One particularly annoying problem had been pestering engineer and machine learning enthusiast Roni Bandini — a neighbor that blasted reggaeton music every morning with the regularity of Old Faithful. Bandini notes that he does not have anything against reggaeton music per se, but when it makes the walls vibrate every day at nine in the morning, it quickly gets old. Most apartment dwellers would either grin and bear it, or get out their trusty old noise-canceling broom to bang on the ceiling. But being a member of the Hackster community, Bandini chose to take a more interesting approach.

After thinking long and hard — and probably considering many systems involving sharks, lasers, and sharks with lasers — Bandini ultimately settled on the idea of building a device called Reggaeton Be Gone. This device leverages machine learning to listen for the specific sound of reggaeton music, then it sends jamming signals to interfere with nearby Bluetooth speakers that are playing the offending music.

The Reggaeton Be Gone is powered by a Raspberry Pi 3 single-board computer, which has sufficient computational horsepower to run an audio classification neural network, and also comes equipped with a Bluetooth transceiver. A USB microphone was also added to the build to capture audio. Bandini finished off the device by 3D printing a slick case with a 128 x 32 pixel OLED display, because if you are going to foil the neighbor’s annoying plans, you should do it with style.

A machine learning classifier was needed that could detect the specific patterns that are characteristic of reggaeton music. Since no pre-trained models could be found by Bandini for this type of music, he collected some audio samples (which the neighbor was freely offering), and uploaded them to Edge Impulse where a neural network classifier could easily be trained and deployed to the Raspberry Pi.

Using this classifier, the Reggaeton Be Gone continually captures audio samples and checks to see if they sound like reggaeton music. If a predefined certainty threshold is exceeded, a Python program jumps into action. It will send a stream of requests to nearby Bluetooth speakers to interfere with their operation. The goal is to either take them offline, or make the audio quality so poor that there is no choice but to turn the music off.

Lots more details are available in the project write-up if you have an annoying neighbor of your own to deal with. It is important to note, however, that jamming Bluetooth devices may not be legal where you live, so be sure to check local laws before you decide to eliminate reggaeton from your own surroundings.