Research On Long Sequence Speech Perceptual Hash Authentication Algorithm Based On Multi-feature Fusion

Perceptual hashing is gaining more and more attention in some multimedia security applications.However,how to balance the robustness and discriminability in speech hashing is still the biggest challenge for hashing algorithms.There are security loopholes in existing speech authentication algorithms from speech acquisition then to data storage to cloud speech hash database.At the same time,the hash sequence constructed in the authentication algorithm is relatively short,and the same hash sequence may belong to different users.This leads to a higher misidentification rate when users are authenticated.Under low signal-to-noise ratio,the matching accuracy of the hash algorithm in the face of complex noise is not ideal.Therefore,this paper makes a study of the above problems.The main research contents and contributions of this paper are as follows:1.In order to solve the problems of low discrimination and poor security in the existing speech authentication algorithms,guarantee the robustness and authentication efficiency of the algorithm,the paper proposes a long-sequence speech perceptual hashing authentication algorithm based on multi-feature fusion and Arnold transformation.Firstly,the algorithm extracts the wavelet logarithmic energy and Mel cepstrum coefficients,constructs the long hashing sequence of the low-frequency coefficients of the two characteristics respectively;then uses the Arnold transform to scramble and encrypt the two sets of hashing long sequences;finally fuses the encrypted sequence for speech authentication.Experimental results show that the algorithm not only has a low error rate,but also ensures the robustness and security of it.Due to the slow efficiency,low matching accuracy of complex noise under low SNR,and the waste of cloud storage resources caused by too long hash sequence in the method.This paper proposes a proposed method,which is speech perception hashing authentication algorithm based on constant Q transform and tensor decomposition.The algorithm first divides the frequency band into sub-bands,and constructs the variance matrix of the subband set;then constructing the feature tensor through constant Q transformation,and reconstructing the target tensor after Tucker decomposition;finally,the target tensor is constructed into a long sequence of binary perceptual hashing to complete speech authentication.Experimental results show that the proposed algorithm is much better than the existing algorithms in discriminability,and simultaneously takes into account both robustness and real-time performance,which meets the requirements of speech authentication in complex noise environment.2.Existing speech authentication algorithms directly make hash structure for the extracted speech features and save them to the cloud,which is easy to cause the leakage of speech features.When constructing a hashing,the utilization efficiency of speech features is poor,and the constructed hashing sequence is short,which will result in insufficient discrimination of the hashing sequence and deviation of authentication.In order to solve the above problems,a long-sequence biometric authentication algorithm based on two-dimensional sine modulation mapping(2D-SIMM)and gammatone filter cepstral coefficient(GFCC)cosine value is proposed.Using 2D-SIMM to construct a biometric security template for the spatial cosine feature of the speech signal,the algorithm verifies the spatial cosine value of three different features.After comparison of simulation experiments,the GFCC algorithm not only reduces the collision rate of biometric sequences and the running time of the algorithm,but also overcomes the impact of content retention operations on the accuracy of authentication.In the face of common low signal-to-noise ratio noise backgrounds,the algorithm also has a good matching accuracy and can also provide a revocable security template for biometrics.