Speech Steganography Based On Adaptive Codebook Partition And Its Detection

Steganography is an emerging information security technology,which can hide information into a multimedia file without causing perceived change.So,it is often used as secret communication.With popularization of network voice technology,steganography based on network voice has become a research hotspot in the field of information security.The pitch period prediction is a key step in speech coding.Because the pitch period is difficult to measure accurately,speech compression coding uses adaptive codebook parameters to approximate it,which afford the opportunity for steganography.Although the steganography can be an effective means of secure communication,it will pose a great threat to network security if used illegally.Therefore,steganalysis research is essential.This thesis focuses on speech steganography based on adaptive codebook partitioning.On the one hand,in view of the problems of weak detection resistance and low hiding capacity of the existing steganography methods,a more secure and effective steganography method is proposed.On the other hand,in order to detect this type of steganography,a more universal and efficient steganalysis method is proposed.The specific research work is as follows:(1)Aiming at the the weak anti-detection ability of traditional speech steganography,a speech steganography scheme based on fractional pitch delay is proposed.Specifically,the research improves steganographic security in three ways.First,the use of decimal parameter has a small effect on voice quality,which improves the sensory transparency of steganography.Second,the introduction of 3 m sequences to encrypt the secret information and guide the steganography process enhances the security of the embedding process.Third,an integer pitch parameter covering strategy is designed to ensure that its steganography remains the same after steganography,further improving the detection resistance.The experimental results show that the proposed scheme has better resistance to detection than the existing schemes.(2)Taking the EVS encoder widely used in 5G voice communication as the carrier object,a speech steganography scheme based on the expandable matrix is proposed.In order to ensure better steganographic transparency,the scheme adopts relative pitch delay parameters and decimal pitch delay parameters,which have less influence on speech quality,as steganographic carriers.In order to further improve the transparency of steganography,a multi-matrix combination embedding method based on convex optimization is proposed.Experimental results show that compared with the traditional matrix embedding method,this method has more flexible embedding efficiency and higher steganographic bandwidth.(3)In order to further improve the detection accuracy of existing steganalysis methods,a pitch delay steganalysis scheme based on parity characteristics is proposed.The scheme combines two types of detection features.One is the proposed parity Bayesian probabilistic(PBP)feature,which can characterize the change of parity characteristics of adaptive codebook parameters before and after steganography.The other is the C-MSDPD feature,which can can characterize the change of the adaptive codebook parameter value changes before and after steganography.Experimental results show that the proposed scheme is feasible and has higher detection accuracy than the existing schemes.(4)In order to enhance the universality and accuracy of speech steganalysis based on adaptive codebook partitioning,after a thorough analysis of existing steganography methods,a steganalysis scheme is proposed.Because the decimal pitch delay parameter is used as the precision of the integer pitch delay parameter during the quantization of the pitch period,there is a strong correlation between the integer parameter and the decimal parameter.The scheme uses the conditional probability of fractional distribution and integer difference distribution as detection features.Based on different types of steganography methods in different coding environments,the performance of the proposed method is evaluated and analyzed.The experimental results show that this method can effectively detect various speech steganography methods based on adaptive codebook partitioning in different coding environments,and can achieve a higher detection accuracy rate than the existing methods.