AC: An audio classifier to classify violent extensive audios

Abstract

This paper presents audio-based classifier to classify the audio into four audio classes like music, speech, gunshots and screams. The audio signals are divided into frames, and various frame-level time and frequency features are calculated for the segment of audio. The classification rules are based on the combination of statistics value calculated for each feature. The classifier takes an unknown segment of audio, applies the classification rules and outputs the label for particular audio. The audio classifier performs with an effective recall rate of 84%.