Speech-Based Biometric Authentication for Secure Smart Home Access

Abstract

As smart home technologies proliferate so quickly, the issue of gaining access to connected environments without losing the convenience of the user becomes a vital concern. The traditional authentication techniques, such as passwords, PINs, and physical tokens, are increasingly being considered inadequate since they can be stolen, spoofed and neglected by users. To this extent, the paper proposes speech-based biometric authentication system that is specifically suited to secure and smooth access of smart home. The suggested solution is advantageous to utilize the individualistic properties of vocal tract features to identify the user by voice, thus giving a non-contact and easy to use option. The use of a hybrid deep learning model (the combination of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks) will allow better representation of spatial and temporal features of speech signals. Mel-frequency cepstral coefficients (MFCCs) as well as spectral and chromatic features are extracted by the system to form sturdy speaker embeddings. These embeddings are then deployed in real time within an embedded system based on Raspberry Pi that makes direct integration with other IoT-based smart home devices (e.g., door locks, lights, and HVAC controls) possible. The verification accuracy of the proposed model is found with experimental validation in the range of 95.8% achieved and an Equal Error Rate (EER) of 2.3% to demonstrate the proposed model has high resistance to spoofing and replay attacks using benchmark datasets, namely VoxCeleb1 and LibriSpeech. Also, the robust performance of the system is stable across different noise levels and it has sub-50 millisecond inference latency, which can be deployed in real-time on edge devices that have limited resources. The given study indicates the possibility of voice biometrics as a secure, efficient, and scalable authentication technique in the smart home setting, providing better privacy and utility than the traditional approaches. The viable potential of using lightweight and secure biometric authentication projects in real life smart living projects is emphasized by the successful combination of deep learning-driven speech recognition and edge IoT structures.