Multilabel Acoustic Event Classification Using Real-World Urban Data and Physical Redundancy of Sensors

Abstract

Many people living in urban environments nowadays are overexposed to noise, which results in adverse effects on their health. Thus, urban sound monitoring has emerged as a powerful tool that might enable public administrations to automatically identify and quantify noise pollution. Therefore, identifying multiple and simultaneous acoustic sources in these environments in a reliable and cost-effective way has emerged as a hot research topic. The purpose of this paper is to propose a two-stage classifier able to identify, in real time, a set of up to 21 urban acoustic events that may occur simultaneously (i.e., multilabel), taking advantage of physical redundancy in acoustic sensors from a wireless acoustic sensors network. The first stage of the proposed system consists of a multilabel deep neural network that makes a classification for each 4-s window. The second stage intelligently aggregates the classification results from the first stage of four neighboring nodes to determine the final classification result. Conducted experiments with real-world data and up to three different computing devices show that the system is able to provide classification results in less than 1 s and that it has good performance when classifying the most common events from the dataset. The results of this research may help civic organisations to obtain actionable noise monitoring information from automatic systems.

Analysis and Acoustic Event Classification of Environmental Data Collected in a Citizen Science Project

“…With that assumption, there are some recent works that are rated with the same metric as this one (event based F1-Score) that can be useful as a reference. A recent previous work from one of the authors [ 62 ] achieved a micro F1-Score of 46% and a macro F1-Score of 12% with a similar dataset of 21 classes of outdoor urban sounds prior to data augmentation. The top ranked work in the DCASE2020 Challenge Task 4 [ 65 ] (DCASE2022 chose a different metric to assess the performance) scored 41.7% in the event based F1-Score before data augmentation using a dataset of ten classes of indoor sounds.…”

“…Although a significant part of the related literature is covered in the aforementioned review, other more recent contributions include some publications related to the last DCASE challenges [ 60 , 61 ], more focused on indoor domestic sounds, and Vidaña-Vila et al’s work [ 62 ]. In this study, the authors developed a polyphonic sound event classifier for urban data using physical redundancy of sensors.…”

Analysis and acoustic event classification of environmental data collected in Sons al Balcó Project

“…These results are consistent with other recent works in the literature that deal with similar complex polyphonic datasets. For comparison, a study conducted in 2021 [15] achieved an instance averaged F1-Score of 46% and a class averaged F1-Score of 12% before data augmentation using a similar dataset of 21 outdoor urban sounds classes. F1-Scores are improved when opting for a segment-based approach.…”

Toward the Definition of a Soundscape Ranking Index (SRI) in an Urban Park Using Machine Learning Techniques

“…In urban areas, ML models have been applied for predicting long-term acoustic patterns from short-term sound pressure level measurements [ 30 ] and for detecting anomalous noise sources prior to computing traffic noise maps [ 31 ]. CNNs have been applied to soundscape classification [ 32 , 33 ], species-specific recognition [ 34 , 35 , 36 ], and the identification of multiple and simultaneous acoustic sources using a two-stage classifier able to determine, in real time, simultaneous urban acoustic events taking advantage of physical redundancy from a wireless acoustic sensors network (WASN) in the city of Barcelona [ 37 ].…”