A Hybrid Iris-Voice Biometric Framework for Robust Identification in the Presence of Ocular Impairments

Accurate and reliable biometric identification is essential in high-security applications such as healthcare, finance, and airport control. However, traditional unimodal systems - relying solely on iris or voice recognition - are prone to environmental and physiological limitations, including issues like ocular injuries and acoustic interference. To overcome these challenges, this paper proposes a hybrid biometric framework that seamlessly integrates iris and voice modalities using an dynamic fallback strategy. The system prioritizes iris recognition, employing a robust segmentation approach based on the Expectation-Maximization (EM) algorithm, which models iris texture through Gamma and Normal mixture distributions. When iris data is incomplete or compromised, the system adaptively transitions to a BiLSTM-CNN based voice recognition module. This dual-layer architecture leverages feature-level fusion to enhance decision accuracy across both modalities. Experimental validation using the CASIA and VoxCeleb datasets demonstrates that the proposed hybrid system achieves an identification accuracy of 96.4%, substantially outperforming unimodal counterparts. Further statistical analysis - including 95% confidence intervals and paired t-tests (p-value = 0.0002) - confirms the robustness and reliability of the approach. Overall, this solution presents a scalable and resilient method for biometric authentication in complex, multimodal environments.