(a) The accuracy of no protection scheme, PPiSV, HESR and AGREE.
(b) The accuracy of AGREE with σ=0,10,15.
About
Maintained by
Qi Li, Hui Zhu, Xiaopeng Yang, Rongxing Lu, Ziling Zhang, and Hui Li
Xidian University
266 Xinglong Section of Xifeng Road, Xi’an, Shaanxi 710126, China
Abstract
With the booming of Internet of Things (IoT) and intelligent terminals, the applications of speaker recognition have been seen increasingly rapid advances in the past two decades. However, the flourish of speaker recognition technology still faces many challenges in IoT scenarios, especially in preventing the disclosure of voiceprint. In this paper, aiming at addressing the challenges, an efficient and privacy-preserving speaker recognition scheme, named AGREE, is proposed for IoT. With AGREE, the speaker identity can be recognized at multiple security levels while without leaking the voiceprint data. To be specific, based on the random matrix theory, a voiceprint encryption algorithm and the corresponding ciphertext-based similarity computation algorithm are proposed. Resting upon these algorithms, our efficient and accurate speaker recognition AGREE scheme can be achieved. Detailed analysis shows that AGREE can resist various known security threats. In addition, AGREE has also been implemented with a real speaker voiceprint dataset, and extensive simulation results demonstrate that AGREE is highly accurate and efficient, and can be flexibly deployed in real environment.
System Model
Overview of AGREE
Accuracy
Recognition computation time in no protection scheme, AGREE, PPiSV and HESR.
Computation overhead of AGREE.