Privacy-preserving PLDA speaker verification using outsourced secure computation

Abstract

The usage of biometric recognition has become prevalent in various verification processes, ranging from unlocking mobile devices to verifying bank transactions. Automatic speaker verification (ASV) allows an individual to verify its identity towards an online service provider by comparing freshly sampled speech data to reference information stored on the service provider's server. Due to the sensitive nature of biometric data, the storage and usage thereof is subject to recent EU regulations introduced as means to protect the privacy of individuals enrolled in an automatic biometric verification system. Stored biometric data need to be unlinkable, irreversible, and renewable to satisfy international standards. Preserving privacy in ASV is also important because, contrary to other biometric characteristics such as fingerprints, speech data can be used to infer a lot of sensitive information about the data subject. As a result, some architectures have been proposed to enable privacy-preserving ASV in the encrypted domain. Recently, homomorphic encryption (HE) was proposed to protect both subject features and vendor models in an embedding-based ASV. This architecture improves on previous privacy-preserving ASV by using (probabilistic) embeddings (i-vectors) and by additionally protecting the vendor's model. However, the usage of HE comes with a rather heavy overhead and significantly slows down the verification process. In this work, we align the cryptographic notion of outsourced secure two-party computation to embedding-based ASV. Our architecture protects biometric information in ASV and can also be used for any automatic biometric verification task. We show that unlinkability, irreversibility, and renewability are granted. Compared to the HE solution, our architecture results in considerably lower communication and computation overhead. Our architecture has been implemented and is experimentally evaluated on the NIST i-vector challenge 2014 using the cosine distance and log-likelihood ratio (LLR) scores from probabilistic linear discriminant analysis (PLDA) and two-covariance (2Cov) comparators. The results show that verification accuracy is retained while e ciency is improved. For instance, a PLDA verification with an embedding dimension of 200 takes about 77 milliseconds over a LAN. This is an improvement of more than 3000⇥ over the HE-based solution and shows that privacy of subject and vendor data can be preserved in ASV while retaining practical verification times. Moreover, our system is secure against malicious client devices.

Analysis of Methods and Techniques Used for Speaker Identification, Recognition, and Verification: A Study on Quarter-Century Research Outcomes

“…Treiber et al addressed this issue and proposed architecture to enable privacy-preserving ASV in the encrypted domain. The authors stated that their proposed architecture is performing better than a previous scheme, namely the homomorphic encryption (HE), where the usage of the latter encryption comes with a rather heavy overhead, leading to a slow verification process [8].…”

Audio-Visual Biometric Recognition and Presentation Attack Detection: A Comprehensive Survey

“…There are ample amount of literature available for face images in cancelable biometrics and in cryptobiometrics [76] also in homomorphic encryption [43], [96], [115]. And also there are limited amount of literature available for speech in the cancelable biometrics [17], [92], [99], [129], [142] and in homomorphic encryption [93], [131]. The privacy-preserving techniques in AV biometrics have not received much attention in the audio-visual domain equivalent to other biometrics.…”

Audio-Visual Biometric Recognition and Presentation Attack Detection: A Comprehensive Survey

“…There are ample amount of literature available for face images in cancelable biometrics and in cryptobiometrics [77] also in homomorphic encryption [44], [97], [116]. And also there are limited amount of literature available for speech in the cancelable biometrics [17], [93], [100], [130], [143] and in homomorphic encryption [94], [132]. The privacy-preserving techniques in AV biometrics have not received much attention in the audio-visual domain equivalent to other biometrics.…”