Design and Analysis of Voice Template Protection Schemes Based on Winner-takes-all Hashing

• Main Author: Chee, Kong Yik
• Format: Final Year Project / Dissertation / Thesis
• Published: 2018
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utar.edu.my/3609/1/SCA%2D2018%2D1601194%2D1.pdf; http://eprints.utar.edu.my/3609/

The emergence of mobile technologies has greatly popularised the usage of speaker recognition based applications (e.g., biometric lock, voice banking and attendance system). However, with the increase in the usage of these biometric based applications, major concerns of the users will be on the security and the privacy of the information stored in the database. Template protection scheme is designed to secure the user biometric from being recovered by attackers. However, current voice template protection schemes do not completely provide the required security properties, such as non-invertibility, unlinkability and revocability. In this dissertation, two voice template protection schemes that are inspired from Winner-Takes-All hashing are proposed. These two newly proposed schemes are named as Random Binary Orthogonal Matrices Projection hashing and two-dimensional Winner-Takes-All hashing. The former scheme is designed for one-dimensional input while the latter scheme is designed for two-dimensional input. To further increase the security of the proposed schemes, additional factor of authentication is incorporated (i.e., random token and additional biometric modality). Extensive analysis is performed to justify the trade-off between the performance and the security of the proposed schemes. The experimental results and analysis have demonstrated that both of the proposed schemes are able to survive against major privacy and security attacks (e.g., attack-via-record multiplicity and stolen token attack) while preserving the performance of the proposed schemes.