Keep It Down Up There!

Living in an apartment can be a great experience, but it can also come with some major downsides. One of the most common issues is noisy upstairs neighbors. This can be a huge source of frustration for people living in apartment complexes, as the noise can be disruptive to everyday life.

The most common noise complaints from upstairs neighbors are loud conversations, music, and the sound of feet stomping around on the floor. This can be especially annoying if it occurs late at night when most people are trying to sleep. It can also be disruptive during the day when people are trying to get things done.

To make matters worse, apartment residents often feel powerless to do anything about it. Simply talking to the upstairs neighbor may not be enough to make them stop, and other forms of confrontation can be difficult or even dangerous.

One option is to contact the landlord or property manager about the issue. They may be able to intervene or take other steps to help reduce the noise. However, it can be difficult to prove that there is a problem — it is one person’s subjective opinion against another's. Moreover, in a large apartment complex, it can also be difficult to conclusively determine which resident is the source of a noise disturbance.

A team of researchers at the Korea Institute of Civil Engineering and Building Technology that are apparently fed up with banging their broom handles on the ceiling to try and quiet their upstairs neighbors down developed a novel solution to the problem. They created a machine learning-powered system that takes subjectivity out of the equation to identify noisy neighbors without playing the he said, she said game.

Likely the first method that comes to mind for most people when it comes to measuring audio levels is a microphone. However, installing microphones in residents’ apartments is highly intrusive and would be unacceptable to virtually everyone. For this reason, the team chose to use vibration sensors installed throughout the flooring. These sensors allow the system to monitor footstep-induced vibration in real-time.

Determining exactly what those vibrations represent, and if they are a nuisance, is another matter entirely. So, to interpret the vibrations, the researchers turned to a machine learning model — a convolutional neural network, specifically. By taking this approach, they were able to sidestep the difficulties associated with employing a traditional statistical approach that can deal with all of the complexities of different building materials and other conditions. It also avoided the hours-long computations required when leveraging physics-based models.

While the model was trained with recognizing footstep sound volume in mind, it can also detect other impact-based sounds like hammering. Validation of the algorithm showed that a mean absolute error of less than one decibel had been achieved when examining two seconds of vibration measurements. To put the sensitivity of this technique into perspective, the sound of a human breathing registers at approximately ten decibels.

This information gives residents and property owners all of the hard, objective data that they need to resolve conflicts. Looking to the future, the team envisions their device being used to directly inform residents when they are being too noisy to remind them to change their behavior, effectively stopping the problem before it becomes a problem.