Study On Active Forensics And Homomorphic Encryption Of Digital Speech

With the development of digital speech communication technology and the wide use of digital speech in social networks,people pay more and more attention to the authenticity of digital speech content and the security of sensitive speech.The emergence of various powerful digital speech editing software makes it relatively easy to tamper with and edit digital speech,which makes digital speech tampering seriously threaten the security of digital speech content.At the same time,in order to meet the needs of large-scale digital speech management,more and more digital speech are stored in remote cloud servers,which makes the security of sensitive speech seriously threatened.Therefore,how to identify the authenticity and integrity of digital speech,how to protect the confidentiality of sensitive digital speech and how to ensure the availability of cipher-text speech have become the valuable research directions in the network security.In this thesis,we focus on the studies of active forensics technology and homomorphic encryption technology of digital speech.The primary contributions are list as follows:1.Perceptual hash and robust watermarking have been widely investigated to solve the problems of authenticating speech integrity.The former generates a watermark and the latter embeds the watermark into the speech signal to implement speech integrity authentication.This paper propose an active forensics algorithm based on perceptual hashing and learned dictionaries.To obtain speech perceptual hash values,we propose a Gammatone filter model of the speech signal to extract stable auditory features(denoted by Gammatone features).A random Gaussian matrix is used to reduce the dimensionality of the features of the Gammatone,then using discrete wavelet transform and vector norm to generate speech perceptual hash values.For the watermarking process,we construct learned dictionaries to obtain the robust sparse feature of coefficients of the stationary wavelet transforms,and embed a watermark(speech perceptual hash values)into the sparse feature by patchwork and quantization index modulation.We illustrate the good performance of the active forensics algorithm in terms of imperceptibility,and embedding capacity,and verify its robustness against common signal processing operations while maintaining imperceptibility.Moreover,our proposed method is sensitive to the malicious modification of the watermarked speech,these malicious modification include mute attack,substitution attack,mixed attack,etc.Compared with other active forensics algorithms,the proposed algorithm can obtain better robustness in the detection and localization of tampering with the content of speech.2.This paper propose an effective active forensics algorithm for encrypted speech signals based on Non-negative Matrix Factorization.The algorithm includes two parts: speech encryption and cipher-text speech active Forensics.In speech encryption section,first,the original speech signal is scrambled,then Integer Wavelet Transform is performed on the scrambled speech signal to obtain the approximation coefficients and the detail coefficients.Afterwards,Advanced Encryption Standard is used to encrypt the approximation coefficients,and finally the encrypted speech is obtained.In cipher-text speech active Forensics section,Non-negative Matrix Factorization is employed on the encrypted approximation coefficients to generate perceptual hashing bits as the watermark,which is embedded into the detail coefficients.In this way,watermarked encrypted speech is obtained.In authentication phase,on one hand,the reconstructed perceptual hashing is generated by the speech to be authenticated based on Non-negative Matrix Factorization.On the other hand,anther perceptual hashing version is extracted from the watermarked encrypted speech.The integrity verification of encrypted speech is conducted by comparing the reconstructed perceptual hashing bits with the extracted perceptual hashing version.Experimental results show that not only our method is sensitive to malicious tampering of encrypted speech signals,but also our method can tolerate common signal processing operations.The comparison shows that our algorithm obtain better robustness and larger embedding capacity than the former exiting methods.3.This paper presents a speech homomorphic encryption scheme with less data expansion,which is a probabilistic statistics and addition homomorphic cryptosystem.In the proposed scheme,the original digital speech with some random numbers selected is firstly grouped to form a series of speech matrices.Then,a proposed matrix encryption method is employed to encrypt these speech matrices.After that,mutual information in sample speech cipher-texts are reduced to limit the data expansion.Performance analysis and experimental results show that the proposed scheme is addition homomorphic,and it not only resists known-plaintext attack and statistical analysis attacks but also eliminates some signal characteristics of original speech.In addition,comparing with Paillier homomorphic cryptosystem,the proposed scheme has less data expansion and lower computational complexity.Furthermore,the time consumption of the proposed scheme is almost the same on the smartphone and the PC.Thus,the proposed scheme is extremely suitable for secure speech storing in public cloud computing.