Neural network classification of gunshots using spectral characteristics

Abstract

The paper describes neural network classification of specific audio sources into given categories. Audio sources are represented by various gunshots, from handguns or big bore guns. This article is a follow-up to an existing system for localization of audio sources from security and military areas. The question of successful classification lies in the convenient discrimination of the feature vector in the feature vector space. A set of feature vectors based on power spectral density is evaluated a tested for the best classification of gunshots.