Voiceprint Challenge for Cryptographic Authentication Based on Triplet Networks and Fuzzy Extractors

Due to its uniqueness and readiness availability, biometric information has been widely used for secure authentication in various contexts, such as user authentication, authorized access, and online payments. However, biometric information involves personal privacy, and the potential leakage of biometric information is extremely harmful. At present, most of the methods that use biometric information for authentication, such as fuzzy extractors, only protect the template, but cannot resist biometric information replay attacks. This paper proposes a voiceprint challenge-based cryptographic authentication framework (VCCAF), which integrates the user’s ownership of the terminal device into one-time voiceprint challenge to overcome possible biometric information replay attacks. The one-time voiceprint challenge constructs a text content with voiceprint information removed through secure multi-party computation, thus avoiding the possible leakage of the raw sound data due to network transmission. Moreover, a time-delay neural network (TDNN)-based triplet network is employed to extract voiceprint features, and then a reusable fuzzy extractor is applied to complete the challenge by passing the local voiceprint check. The experimental results show that the proposed framework supports incremental registration of users with excellent scalability, and provides a promising cryptographic authentication paradigm for practical authorized access applications.