Traditional machine learning-based steganalysis methods on compressed speech in VoIP applications have achieved great success. However, in these methods, there is a dilemma between the effectiveness of modeling the steganographic carrier and the high dimensionality of extracted features. Especially for small-sized and low embedding rate samples, most existing methods do not perform well enough. To deal with this issue, we present MDoIF— an Adaptive Multi-Rate (AMR) steganalysis of compressed speech based on multi-domain information fusion. In order to fully extract the information reflecting the change of carrier correlation before and after VoIP steganography, we construct a Bayesian network with FCB parameters in compressed speech as the vertices, and quantify link strength between codebook parameters. On this basis, we design a multi-domain feature extraction algorithm, supplemented by an information-theoretic measure-based feature selection algorithm for dimensionality reduction, which can significantly improve the performance of MDoIF. To evaluate the performance of our method, we conduct comprehensive experiments on MDoIF and existing models. Experimental results show that MDoIF performs effectively on various AMR steganalysis tasks with excellent detection accuracy. Particularly for small-sized and low embedding rate samples, MDoIF surpasses the state-of-the-art methods.

Traditional machine learning-based steganalysis methods on compressed speech have achieved great success in the field of communication security. However, previous studies lacked mathematical modeling of the correlation between codewords, and there is still room for improvement in steganalysis for small-sized and low embedding rate samples. To deal with the challenge, we use Bayesian networks to measure different types of correlations between codewords in linear prediction code and present F3SNet—a four-step strategy: embedding, encoding, attention, and classification for quantization index modulation steganalysis of compressed speech based on the hierarchical attention network. Among them, embedding converts codewords into high-density numerical vectors, encoding uses the memory characteristics of LSTM to retain more information by distributing it among all its vectors, and attention further determines which vectors have a greater impact on the final classification result. To evaluate the performance of F3SNet, we make a comprehensive comparison of F3SNet with existing steganography methods. Experimental results show that F3SNet surpasses the state-of-the-art methods, particularly for small-sized and low embedding rate samples.