What is noise and how do we measure it? Using machine learning and an inexpensive Internet of Things unit, researchers at Linnaeus University have developed an approach for noise classification.
Noise is any undesired environmental sound. And not only can it cause annoyance, it may also negatively affect your health.
A European Union directive stipulates that all member states must assess environmental noise and develop noise maps every five years. But that is far from sufficient, since noise may change over time. Also, health-damaging noise often occurs during short time periods only. And the type of noise – music, traffic, rumbling from construction sites etc – affects how you perceive it.
Identify the type of noise
In other words, to get a true idea of noise pollution you need to not only measure the dB level, but also identify the type of noise. To solve this challenge, researchers at Linnaeus University have developed an approach for noise classification in smart cities using machine learning on an inexpensive Internet of Things unit. The unit comprises a Raspberry Pi Zero W and a dual-microphone expansion board. Raspberry Pi is a single-board computer the size of a credit card, costing only about 10 Euro.
To investigate the performance of the system, the researchers conducted experiments with eight different classes of environmental sounds: quietness, silence, car horn, children playing, gunshot, jackhammer, siren, and street music. To train it, they used a dataset with more than 3,000 sound clips.
Noise pollution monitoring in real-time
“We observed that our solution provides high noise classification accuracy, in the range of 85–100 %. In the future, we plan to use our devices for large-scale noise pollution monitoring in real-time. Devices like this one may be placed at many points of interest across a city, and the data would be accessible in open form to citizens and to health and environment protection units of the city”, says Sabri Pllana, senior lecturer of computer science at Linnaeus University and team manager.
The team’s work has been described in the new issue of HiPEAC magazine (January 2019, issue 56, page 29). HiPEAC is short for High Performance and Embedded Architecture and Compilation, a European network of 2,000 world-class computing systems scientists, industry representatives and doctoral students.