A Cancelable Biometric Approach for Efficient Identification of Speakers from Encrypted Speech

Biometric systems based on feature extraction from speech signals are extensively deployed in different security systems. This work presents a secure speaker identification system, namely cancelable speaker identification system, which can be further used for remote access applications. Two formulations for efficient cryptosystems based on a chaotic map and the empirical mode decomposition (EMD) with a single key are presented for this purpose. The encryption achieves a large degree of security in the speaker identification systems. In the proposed cancelable speaker identification scenario, cepstral features are extracted from encrypted speech signals. The matching in this proposed scenario is performed with artificial neural networks (ANNs). Furthermore, various enhancement methods are implemented at the receiver side to remove the noise effect prior to speaker identification. Consequently, the features of the speakers are protected from the attackers. The simulation results prove that the proposed cancelable biometric system has an outstanding performance level compared to the traditional biometric systems. It achieves a recognition rate up to 100% at a − 20 dB signal-to-noise ratio ( SNR ) and above, using the two proposed cryptosystems with a high level of security.