Stadtlärm - a distributed system for noise level measurement and noise source identification in a smart city environment
Various types of acoustic scenes such as construction sites, open air concerts, sport events, and road traffic contribute to the overall rising noise pollution in cities. As part of the ""Stadtlärm"" research project, we propose an intelligent audio analysis system that is implemented on distributed sensors in a smart city network. The system allows for measuring noise levels according to the general administrative regulation ""TA Lärm"" as well as classifying and localizing sources of active acoustic events. Results of this measurement can be visualized online or collected for the further analysis. The local city administration is supported in processing incoming resident noise complains by continuously monitoring the noise distribution across the city. Furthermore, local noise appearance can be predicted for planned events in the future by correlating previous measurements with metadata and geo-locations obtained from event calendars. On each sensor, Deep Convolutional Neural Networks are used for automatic feature learning from spectrogram data and the joint classification of 12 possibly concurrent acoustic event classes and 5 mutually exclusive acoustic scene classes. We perform a systematic analysis of the influence of various model hyperparameters as well as the presence of additional environmental background noise to the classifier performance.