Do you want to know how much noise there is on a construction site? You want to measure the sound of the excavator or the hammer drill, but you don’t want to measure seagulls, traffic noise or a helicopter flying by.

Now a new sound measurement service can do just that – with the help of artificial intelligence. But that intelligence is not a given, says Femke B. Gelderblom, a SINTEF senior research scientist.

“Sound and artificial intelligence research is still quite an immature field when compared to what AI can do with images or text. Audio is very difficult, even though it is an easy concept for us humans to understand. Explaining that a certain sound comes from the same source as a similar sound, albeit from a greater distance, can quickly confuse an AI model!” Gelderblom says.

But Gelderblom and her colleagues at SINTEF have figured it out. They collaborated on a research project with Norsonic, a Norwegian company that makes equipment for measuring sounds and vibrations. The result has been a completely new product called NoiseTag. “It’s exciting to be able to do research on topics that turn into real products,” she says.

“The challenge up to now has been that people who go through the data have to listen to each individual recording to remove the measurement data that is not relevant to the project,” says Karl Henrik Ejdfors at Norsonic.

“Seagull cries at a harbour are a typical example. Instead of listening to seagull cries and helicopters, we train a model on these sounds and identify which sounds should be included and which should not be included in the project. Then the model automatically filters and removes those parts of the measurements,” he says.

Norsonic produces measuring equipment, including for ports and construction sites. The microphones are calibrated and standardized to meet the Norwegian statutory requirements for noise measurement, including in the construction industry. But the microphones pick up all the sound on site.

“Discerning what is relevant noise and what isn’t is key when monitoring sounds. A construction project is very often located next to a major road. The road is noisy, but that doesn’t mean the construction project shouldn’t be built because all the sources of noise make the construction site too noisy. That’s when you need to know what noise comes from the road, and what noise is from the construction project itself,” says Gelderblom.

Research has made it possible to create models that are specific to each individual customer. This technology can distinguish the noise that each customer wants to measure.

“Humans can learn this quite quickly. I recognize what I hear. Machines can also learn to do this, but it’s not as easy for them,” she says.

Things like the surroundings and distance affect the sound. The sound from the source changes before it reaches you. In addition, the sound can vary even if your ear hears that it has a corresponding source.

For example, one excavator may have a different engine speed than another one does. For you and me, it is easy to hear that both sounds are from an excavator, but for a machine that is listening, they are two different sounds, and it has to learn that both are being produced by an excavator.

NoiseTag has just been launched. “Typical of artificial intelligence is that it gets smarter the more we use it and the more data comes in. We’re hoping that our device will get better as more people use it,” says Gelderblom.

SINTEF is now building on NoiseTag with a new research project in collaboration with Norsonic and NINA. The Robust Acoustic Recognition (RoAR) research will minimize the manual work required to set up a new system for classifying new types of sound sources for AI.

The RoAR project will continue until 2028. The Research Council of Norway and Norsonic are funding the project.