Discrete Wavelet Transform Based Cancelable Biometric System for Speaker Recognition

Abstract

The biometric template characteristics and privacy conquest are challenging issues. To resolve such limitations, the cancelable biometric systems have been briefed. In this paper, the efficient cancelable biometric system based on the cryptosystem is introduced. It depends on permutation using a chaotic Baker map and substitution using masks in various transform domains. The proposed cancelable system features extraction phase is based on the Cepstral analysis from the encrypted speech signal in the time domain combined with the encrypted speech signal in the discrete wavelet transform (DWT). Then, the resultant features are applied to the artificial neural network for classification. Furthermore, wavelet denoising is used at the receiver side to enhance the proposed system. The cryptosystem provides a robust protection level of the speech template. This speech template can be replaced and recertified if it is breached. Our proposed system enables the generation of various templates from the same speech signal under the constraint of linkability between them. The simulation results confirmed that the proposed cancelable biometric system achieved higher a level of performance than traditional biometric systems, which achieved 97.5% recognition rate at low signal to noise ratio (SNR) of -25dB and 100% with -15dB and above.

Keywords

• biometric system
• cancelable biometric system
• mel-frequency cepstral coefficients
• discrete wavelet transform
• wavelet denoising
• speech encryption
• chaotic baker map