Detecting Steganography in AMR Speech Based on Pulse Correlation

Abstract

This paper presents a novel methodology to detect the steganography on the fixed codebook (FCB) of adaptive multi-rate (AMR) speech stream. We have found that correlations of pulses are influenced by the steganographic operation. Based on this, two categories of features are proposed to characterize the pulse correlations, namely subframe-level pulse correlation based on self-information and track-level pulse correlation based on mutual-information, whose feature dimension is only 1/5 of the state of the art. The proposed method employs the support vector machine as the classifier and is evaluated with a large quantity of AMR speech samples. The experimental results demonstrate that the propose method is effective and has a better detection performance than the state of the arts.