Active Learning Based Audio Tampering Detection

Abstract

Audio authentication is the primary task in an audio forensics scenario in which audio tampering detection is one of the objectives. In this paper, we offer a fresh approach to audio tampering detection using supervised learning and active learning methods. The present techniques are based on supervised learning, and they require a massive amount of labeled data for classification. There is very little availability of standard data. The paper provides a comparative study of supervised and active learning approaches. The work uses unlabeled dataset for classification which is the primary focus in any active learning method. The proposed work uses less than 1-sec audio files for copy and move tampering. Result gives 92.78% accuracy for supervised learning using stft whereas for active learning it gives 87.38%. Active learning reduces the cost of annotation as we do not have to label all the data.