Steganography is a promising technique for covert communications. However, illegal United States of Americage of this technique would facilitate cybercrime activities and thereby pose a great threat to information security. Therefore, it is crucial to study its countermeasure, namely, steganalysis. In this paper, we aim to present an efficient steganalysis method for detecting adaptive-codebook based steganography in adaptive multi-rate (AMR) speech streams. To achieve this goal, we first design a new low-dimensional feature set for steganalysis, including an improved calibrated Markov transition probability matrix for the second-order difference of pitch delay values (IC-MSDPD) and the probability distribution of the odevity for pitch delay values (PDOEPD). The dimension of the proposed feature set is 14, far smaller than the feature set in the state-of-the-art steganalysis method. Employing the new feature set, we further present a steganalysis scheme for AMR speech based on support vector machines. The presented scheme is evaluated with a large number of AMR-encoded speech samples, and compared with the state-of-the-art one. The experimental results show that the proposed method is effective, and outperforms the state-of-the-art one in both detection accuracy and computational overhead.

steganalysis, steganography, adaptive multi-rate speech, pitch delay, support vector machines