AN AUTOMATED GUNSHOT AUDIO CLASSIFICATION METHOD BASED ON FINGER PATTERN FEATURE GENERATOR AND ITERATIVE RELIEFF FEATURE SELECTOR

Abstract

Audio forensics applications and methods are very crucial to clarify crimes. To accelerate audio analysis process and classify audios with high accuracy, machine learning (ML) methods must be used in audio forensics. An automated gunshot audios classification method is presented in this study. To implement our automated gunshot classification method, a novel gun audios dataset was collected from YouTube with 8 classes in the first phase. A novel ML method is presented in the second phase and the proposed ML method contains three fundamental phases. These phases are a novel finger pattern (finger-pat), statistical moments and discrete wavelet transform (DWT) based feature generation network, informative/distinctive feature selection with iterative ReliefF (IRF) feature selector and classification with a k nearest neighbors (kNN) classifier (shallow) to show success of the generated and selected features by using the proposed finger-pat based feature generation network and IRF feature selector. These methods and kNN achieved 94.48% classification accuracy. These results demonstrate that our proposed method can be used in gunshot audio analysis.