Automatic Identification of city intrusive noises using Deep Learning

Long-term noise monitoring is a powerful and useful approach for assessing the acoustic impact of a construction site on the neighbourhood. Its application may also help on evaluating the indoor noise of places such as schools, hospitals, and offices. In most cases, the site under inspection is located in cities, which is inherently prone to intrusive noises, especially those coming from traffic. Identifying intrusive noise for each situation is a challenging task for machine learning algorithms, and also a time saving procedure on long-term noise monitoring. This paper is about the creation of an audio dataset for some typical city intrusive noises, and its application on a low-cost monitoring system called OTOH. It will also be discussed some methods of automatically determining the acoustic labelling of each part during a long-term monitoring using Deep Learning algorithm.