Development and validation of an anomalous noise events detector focused on salient events through an urban and suburban wasn adapted to real-operation

Environmental noise has become one of the biggest health problems facing cities nowadays, being traffic the main noise source that affects their inhabitants. For this reason, the LIFE DYNAMAP project has developed a Wireless Acoustic Sensor Network (WASN) to monitor Road-Traffic Noise (RTN) levels dynamically. To that effect, those Anomalous Noise Events (ANEs) unrelated to regular traffic (e.g., sirens, street works, etc.) should be detected and removed from the RTN maps computation to avoid undesired biasing. To that effect, an Anomalous Noise Event Detector (ANED) has been designed as a two-class classifier (RTN or ANE) to run locally on all the low-cost sensors of the network. After deploying a WASN in pilot area of the project, this work presents the evaluation of the ANED algorithm adaptation to real-operation in both urban and suburban areas, after being trained using more than 310 h of WASN-based collected data. The performance of the algorithm is validated on 87 h of data with ANEs whose subjective salience has been manually labelled by experts, focusing on those ANEs entailing a high salience with respect the surrounding RTN events. The results show that the ANED obtained encouraging accuracy detection rates for the high-salience ANEs (67-78%) in both pilot areas, with more than 90 % of RTN accuracy detection, which minimizes the false alarm probability.