Research On Privacy Protection In Speaker Recognition Based On Homomorphic Encryption

With the popularization of smart devices and the rapid development of technology,traditional identity verification methods,such as ID cards,passwords,and signatures,can no longer meet people's requirements for data security and information protection.At present,most biometric systems directly store the user's characteristic information or authenticate the user's identity,which is easy to be attacked,which leads to the leakage of user information.Therefore,constructing a safe and reliable biometric identification scheme is a hot issue that people pay attention to.Among them,building a safe data template and scoring standard becomes the key.Compared with other biological characteristics,voiceprint recognition has the advantages of non-contact,large amount of information,and low cost.It is an identity authentication technology with good practical value and research potential.Therefore,this article chooses to protect the privacy of voiceprint recognition technology.The main work of this paper has the following aspects:(1)Regarding the security and privacy of the characteristic information of voiceprint recognition,this paper designs a voiceprint template protection method based on random mapping technology.The method is based on identity vector(i-Vector)and linear discriminant analysis(LDA),and proposes an improved random mapping technology.Through the improvement of the random mapping algorithm and the processing of voiceprint feature information,a template protection scheme for voiceprint recognition is constructed.Users can register in the random domain and complete the voiceprint recognition.We have carried out experimental simulations on the proposed method through a known Chinese data set.The experiment shows that the method basically does not affect the accuracy of voiceprint recognition,and it can also effectively ensure the security of voice data.(2)On the basis of ensuring the feature information of the voiceprint,the security of the scoring standard during the authentication phase has become our main concern.This paper proposes the privacy protection of the cosine discriminator based on secure two-party calculation.First,extract the user's identity feature and normalize it,and then combine it with the cosine similarity through a secure two-party calculation protocol to achieve the purpose of protecting the discriminator.This solution also relatively reduces the amount of calculation and time consumption,and has certain restrictions on the access of malicious voices.