Covert Channel Detection Techniques For Instant-Voice Communications

With the extensive application of instant voice,hidden communication based on instant voice has also become a hot issue.Instant-voice-based steganography methods have great transparency because of its instantness,dynamics,and diversity.However,like most security technologies,instant-voice-based steganography is a double-edged sword.Illegal use will bring great security threats.Therefore,the steganalysis based on instant voice is also an important research topic.From the existing researches,there are still some problems.For example,there are no effective steganalysis method for some steganographic methods,and the existing detection methods are based on the premise that the embedded rate and embedded methods are precisely known.To address these concerns,we propose a bitrate modulation based steganalysis method,three steganalysis methods with unknown embedding rate,and a steganalysis method with unknown embedding method,which can be listed as follows:First,as the speech codecs with multiple bitrates can adaptively adjust the bitrate according to the current network conditions,the bitrate switching based steganography method can effectively resist the traditional steganalysis methods.However,the covert bandwidth of this method is small,which can be further enhanced.Thus,we propose two adaptive bitrate modulating based steganographic methods by introducing matrix encoding(ME)strategy.This type of steganographic methods are known as bitrate modulating based steganographic methods.Since they do not modify speech content and protocols,these methods can well resist the traditional steganalysis methods.Therefore,we specifically propose a steganalysis method based on statistical characteristics of the bitrate intervals for these steganographic methods.We use the probability distribution of the bitrate intervals as the detection features,and reduce the dimensions by the detection window and principal component analysis(PCA).The results demonstrate that the proposed method can effectively detect the existing three bitrate modulation based steganographic methods,even at relatively low covert bandwidths.Second,the existing speech-based steganalysis methods assume that the embedding rate of a given test sample is precisely known.However,in practice,we often cannot determine whether the tested sample has been steganographed,let alone the specific embedding rate.To address this concern,we propose three steganalysis methods with unkown embedding rate.Among them,the first two methods are evolved from the existing image steganalysis methods,using different global classifiers.In addition,we also propose a steganalysis method based on Dempster-Shafer's evidence theory(DST).Through comprehensive experiments,the three steganalysis methods compared with the specialized classifier can achieve similar detection effect.Third,to eliminate the assumption of known steganographic methods in the existing steganalysis methods,we propose a steganalysis method with unknown embedded methods.Specifically,first we classify the steganography methods based on adaptive multi-rate speech.Then we train a combined classifier based on DST for each class through a support vector machine.Finally the final classification result is merged with the results of each class according to a pre-defined rule(i.e.,the ouput is positive when the inputs are all positive,otherwise negative).The experiment results with seven kinds of steganographic methods and four kinds of detection features show that the proposed steganalysis method with unkown embedding method can achieve the similar detection effect in comparsion with the specialized classifiers.