Monitoring, profiling and classification of urban environmental noise using sound characteristics and the KNN algorithm. (November 2020)

• Record Type: Journal Article
• Authors: Tsalera, Eleni; Papadakis, Andreas; Samarakou, Maria
• Is Part Of: Energy reports. Volume 6(2020)Supplement 6
• Page Start: 223
• Page End: 230
• Publication Date: 2020-11
• Subjects: Urban noise -- Sound classification -- Supervised machine learning -- KNN algorithm; Power resources -- Periodicals; Energy industries -- Periodicals; Power resources; Periodicals; Electronic journals; 621.04205
• DOI: 10.1016/j.egyr.2020.08.045
• Languages: English
• ISSNs: 2352-4847

Abstract

Abstract: Environmental noise is a key factor affecting the quality of life in modern societies as they influence an extended set of human activities. Unwanted sounds, typically characterized as noise, can be of many types and vary in their impact and the ways to be confronted on behalf of competent public authorities. In this work, we describe environmental noise in a qualitative manner using sound-specific features from the time and spectral domains. These features consist of 8 temporal (including RMS, standard deviation, Zero Crossing Rate), 11 spectral (including spectral envelope slope, skewness, spectrum mass center, peak amplitude crest, spread and skewness) and 4 perceptual (including Mel Frequency Cepstral Coefficients) descriptors. Based upon a set of 8 discriminant types of unwanted sounds, typically met in urban environments (car horn, children playing, dog barking, drilling, engine idling, jack hammer, siren and street music), we specify a methodology of matching environmental noise into these categories. Using training and test data from the UrbanSound8K public dataset, we use the K-Nearest Neighbors (KNN) algorithm for classification. The algorithm has been configured to allow from 1 to 3 neighbors, while three distance metrics (Euclidean, Chebyshev and cosine) have been employed to create 9 models that achieve performance between 70% and 85%.