A Novel Car Interior Sound Classification Method based on Multileveled Local Binary Four Patterns and Iterative ReliefF

Abstract

Sound classification is one of the crucial study areas in machine learning and sound forensics. However, there are limited studies on sound forensics or sound-based crime investigations in the digital forensics literature. In this work, a novel area of sound forensics is presented: car interior sound classification (CISC). The main aim of CISC is to identify a car using its interior environmental sound. A car interior sound dataset was collected using 10 car models. This CISC model includes feature generation using the local binary four pattern and one-dimensional multilevel discrete wavelet transform (DWT), iterative ReliefF-based feature selection, and classification. k-nearest neighbors (kNN) and support vector machine (SVM) were utilized as classifiers to demonstrate the general success of the proposed learning model for CISC. The accuracy rates were calculated as 93.72% ± 0.37 and 95.04% ± 0.30 with kNN and SVM, respectively. These results demonstrate the success of the proposed method.