View article

This paper presents an effective blind speech steganalysis based on entropy features. To detect changes due to steganographic algorithms, each speech signal is divided into four energetic parts using active speech level (ASL) algorithm, defined in ITU-T Recommendation P.56. Maximum entropy is computed from each energy part to generate a set of features fed to a nonlinear SVM classifier with an RBF kernel to distinguish between cover and stego speech signals. Experimental results show that the proposed features are highly sensitive to the change made by the embedding process. The results also reveal that our method performs very well and achieves detection rates up to 98% of stego-signals produced by S-tools4, Steghide and Hide4PGP.