Characterization of a WASN-Based Urban Acoustic Dataset for the Dynamic Mapping of Road Traffic Noise

Abstract

Road Traffic Noise (RTN) is one of the main pollutants in urban and suburban areas, negatively affecting the quality of life of their inhabitants. In the context of the European LIFE DYNAMAP project, two Wireless Acoustic Sensor Networks (WASN) have been deployed to monitor RTN: one in District 9 of Milan, and another along the A90 motorway of Rome. Since the dynamic mapping system should be able to identify and remove those Anomalous Noise Events (ANEs) unrelated to regular road traffic (e.g., sirens, horns, speech, and doors), an Anomalous Noise Event Detector (ANED) has been included in the dynamic noise mapping pipeline to avoid biasing the computation of the equivalent RTN levels. After deploying the 24 low-cost acoustic sensor networks in both pilot areas, WASN-based acoustic datasets were built to adapt the previous version of the ANED algorithm to run in real-operation conditions. In this work, we describe the preliminary results of the analysis of the 154 h WASN-based urban acoustic dataset obtained from the Milan city in terms of the main characteristics of ANEs. The results confirm the unbalanced nature of the problem (83.7% of the data corresponds to RTN), showing the urban WASN-based dataset a larger number of ANEs with higher local predominance than what was observed in the previous expert-based recording campaign, which underlines the importance of the accurate modeling of the urban acoustic environment to train the ANED properly.

Aggregate Impact of Anomalous Noise Events on the WASN-Based Computation of Road Traffic Noise Levels in Urban and Suburban Environments

“…This section describes the results of the experiments from the impact analysis conducted on the two environmental WASN-based audio databases from the DYNAMAP's Milan and Rome pilot areas [39,40]. According to the project specifications, the considered integration time to update the L Aeq,T values of the RTN maps is 5 min [30], i.e., T = 300 s. To analyze to what extent the collected ANEs from each sensor location bias the L Aeq,300s measurement, the impacts are categorized within three impact ranges (i.e., N R = 3) [37], accounting for those occurrences (from either individual or aggregate ANEs) causing a low-impact in θ 1 = (−∞, 0.5) dB, a medium-impact in θ 2 = [0.5, 2) dB, and, finally, a high-impact in θ 3 = [2, +∞) dB, θ 3 = θ c being as this last interval collects those cases that surpass the critical threshold γ c = 2 dB according to the WG-AEN [14].…”

“…The analysis of those datasets showed the highly local and unpredictable nature of anomalous noise events, which were manually labeled and used to train the preliminary version of the ANED algorithm [36]. Recently, the deployment of the two WASNs in both pilot areas has led to the generation of a suburban acoustic dataset through the 19-nodes WASN in Rome [40], together with the completion of the first steps of the creation of an urban dataset through the 24-node WASN installed in Milan in real operation [39]. From these two experiences, it can be concluded that the evaluation of the acoustic salience of any environmental acoustic event is relevant in order to improve the accuracy of the derived machine listening approaches [43], an issue that was justified in [37] after evaluating the individual impact of the detected events on the overall equivalent noise level computation considering 9 h of real-life acoustic data collected through an expert-based recording campaign.…”

“…The audio recordings were collected in continuous raw audio clips from the first 20 min of each hour (considering a sampling frequency of 48 kHz) , as a trade-off between the storage capacity and communications resources of the nodes, and obtaining a representative sub-sampling of the L Aeq measurements along the day [40]. The gathered acoustic data were manually labeled by experts in audio signal processing (see [39,40] for further details). As a result, up to 28 ANE subcategories were identified.…”

“…Besides analyzing the individual impact of ANEs on the measurements, the analysis methodology focuses on evaluating the bias caused by the presence of several ANEs within a given period of time, taking into account their impact range (low, medium or high) and sensor location. The study is conducted on 304 h and 20 min of WASN-based labeled acoustic data collected through both sensor networks, before proceeding to update the ANED algorithm with both WASN-based datasets (see [39,40] for a detailed description of the general characteristics of the urban and suburban datasets, respectively).…”

A Comparative Survey of Feature Extraction and Machine Learning Methods in Diverse Acoustic Environments

“…The goal of this paper was to offer a more general approach to evaluate the performance comparison of all the implemented algorithms over the proposed scenarios. It is known from previous works that the training and test of machine learning algorithms varies substantially [ 28 ] when going to the simulations in the laboratory with a small corpus into real-operation environment, where larger corpora are used [ 29 ], and several unpredictable events can occur. This work faced only the first part of the tests, near a proof of concept of the best combination of FE and ML for each type of group of sounds.…”