Research On Cloud Anonymization Technology Of User Data In Voice Assistant

Voice assistant terminals represented by smart speakers have been loved by consumers in the market in recent years,bringing great convenience to people’s lives.However,the voice assistant terminal also collects the user’s voice data.On the one hand,voice data not only contains the speaker’s voice biometric information,but also may contain private information such as personal emotions,interests and hobbies.On the other hand,many actions of service providers are opaque to users,and these actions may include privacy invasion or privacy leakage.Therefore,the privacy protection problem of users’ voice data needs to be solved urgently.Most of the existing user voice data privacy protection methods are to perform privacy protection processing on the voice data before uploading the voice data to the cloud service,which will affect the user experience.In addition,most methods only focus on one type of privacy contained in speech data,such as emotional privacy,and cannot effectively protect multiple types of privacy contained in speech data.To this end,this dissertation conducts the following research:(1)For privacy protection of user voice data,pseudonymization technology is introduced into the voice assistant scenario to realize the anonymity of user voice data.Under pseudonym technology,device terminals use different pseudonyms to access cloud services and upload user voice data during interaction with cloud services.Cloud services cannot link voice data to real users,and user privacy can be effectively protected.(2)Aiming at the anonymous authentication problem between the device terminal and the cloud service,the identity authentication protocol based on the chameleon hash is applied,which realizes the rapid authentication between the device terminal and the cloud service,and considers the actual computing power of different communication entities.Based on the protocol,an interaction improvement is proposed,which significantly reduces the computation time.And the improved protocol is proven to be secure under the random oracle model.(3)Aiming at the problem that users will leak the relationship between individuals and voice data and between voice data to cloud services when retrieving anonymized voice data,an anonymous security retrieval based on oblivious transfer is proposed.In anonymous secure retrieval,the cloud service sends multiple user voice data including the searcher’s voice data to the searcher.The searcher can only obtain his own voice data,while the cloud service cannot distinguish which voice data belongs to the searcher.This effectively protects the privacy of the searcher’s voice data.The work of this dissertation can effectively protect the privacy of user voice data and meet the actual user’s retrieval needs for personal voice data.In the anonymization of user data,the authorization,authentication and replacement process of pseudonym are all before the interaction between the device terminal and the cloud service,so the anonymization method in this dissertation will not affect the time delay of the interaction process,so it is very practical.In anonymous security retrieval,the oblivious transfer is based on elliptic curve public key system,which can effectively utilize system parameters and reduce system overhead as a whole,so it has important practical value.