<i>SipDeep:</i> Swallowing-Based Transparent Authentication via Bone-Conducted In-Ear Acoustics

The growing use of smart devices requires improving privacy and security. Conventional biometrics confront false positives and unauthorized access, stressing cautious user input. We enhance security by analyzing distinctive human physiological characteristics rather than relying on conventional methods susceptible to spoof attacks. Drinking, a common physiological activity, can provide continuous authentication. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SipDeep</i>, proposed innovative system, utilizes bone-conducted liquid intake sound, incorporating unique biometrics from bone and pharyngeal characteristics. The system captures these elements in the external auditory canal, offering a novel transparent authentication applicable to a diverse user range. Our noise filtering system eliminates environmental and anatomical interferences during drinking, including subtle body movements. The study introduces a hybrid event detection technique integrating wavelet transform with start/end points detection. Next, we extract physiological features from bone structure, liquid intake sound, and liquid intake pattern. We used the physiological features to train a deep learning algorithm based on a Triplet-Siamese network to classify authentication. The proposed model has been thoroughly compared with advanced models such as DenseNet169, ResNet18, and VGG16. Following extensive experimentation involving multiple users across various environments, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SipDeep</i> demonstrates 96.5% authentication accuracy, coupled with a 98.33% resistance to spoof attacks.