Biometric authentication has emerged as a leading approach for the fast and reliable identification and verification of individuals. This work centers on biometric systems with a primary focus on speech signals, a vital biometric trait. Biometric identification relies on distinct physiological or behavioral characteristics, including modalities such as voice, signature, iris, fingerprint, and facial features. Due to the irreversible nature of biometric patterns particularly those derived from speech ensuring confidentiality is crucial. To address this, this work proposes a novel framework for generating reusable and modifiable speech signal templates using cancelable biometric techniques. The proposed framework begins with the application of the Wavelet Scattering Transform (WST), which introduces a non-invertible transformation to the speech signals. This initial step is essential in preventing the reconstruction of the original voice signal, aligning with the key objective of cancelable biometrics: irreversibility. Subsequent to the WST, RSA-based encryption is applied, incorporating user-specific patterns to provide a robust layer of security. The integration of cryptographic techniques with non-invertible transformations significantly strengthens the protection of the cancelable biometric features. The effectiveness of the proposed system is rigorously assessed using performance indicators such as the Equal Error Rate (EER) and the Area Under the Receiver Operating Characteristic Curve (AROC). Experimental results highlight the system resilience and accuracy in the presence of white Gaussian noise and car noise. With an accuracy reaching 99.87% and an EER of 0.0179 for AWGN (TIMIT dataset) and accuracy reaching 99.94% and an EER of 0.0062 for (Chinese Mandarin Corpus). In case of speech corrupted with car noise the accuracy reaches 99.87% and EER of 0.0127 for the TIMIT dataset and 99.85% and 0.0055 for the Chinese Mandarin Corpus, the findings underscore the system's strong potential for secure and reliable biometric authentication.
Kabel et al. (Fri,) studied this question.