Real-time noise event detection using a general exceedance algorithm

Sammanfattning: With urbanization being a global trend, more and more people are finding them- selves living in close proximity to each other. This results in an increase of noise levels due to both the people and the increase of traffic in these areas. Health is- sues as a result of the increased sound levels have become a concern for the cities and governments. Noise events are helpful with identifying causes of disturbance and validating reported complaints. Noise events are in this thesis subjectively de- noted as a noticeable temporary increase in sound level. Usually something that disturbs the exposed individuals. The thesis will introduce a general exceedance algorithm that is constructed to detect and register these noise events in real-time. A general exceedance algorithm works by monitoring the current sound level and when it exceeds a threshold level an event is detected. Typically, in an urban envi- ronment, the background sound level changes, presenting a different sound scene depending on the day and time. This then requires a moving baseline that corre- sponds to the current sound scene. Using a memory of all the sound levels for a set time up to the current time the 50th and 90th percentile (denoted L50 and L90 respectively) can be used as a baseline. Then a threshold is required to make sure only the noise events are captured. The results show that this sort of algorithm is feasible and works on a single board computer, in this case a Raspberry Pi 3, in real-time. Further conclusions include that the LA90 , i.e. the L90 level with the so-called A-weighting filter, performed better than LA50, i.e. the L50 level with the so-called A-weighting filter, and that grouping noise events that occur within short time intervals provided a more accurate result.