RFID Based Smart Mask for Speech Recognition

Lip-reading has become an increasingly significant area of research, aiming to recognize speech through the analysis of lip movements. Predominantly, lip-reading technologies rely on camera-based systems or wearable devices, which face well-known limitations, including occlusion, sensitivity to lighting conditions, privacy concerns, and the discomfort associated with wearable devices that can disrupt daily activities. These issues are further compounded by the widespread use of face masks during the COVID-19 pandemic, which diminishes the effectiveness of vision-based and wearable technologies for speech recognition. To overcome the inherent challenges of these systems, this study proposes an RFID-enabled smart mask as an alternative framework for lip-reading, capable of accurately capturing lip movements beneath face masks and enhancing conversational accessibility for individuals with hearing impairments. By leveraging RFID technology, this system enables RF sensing-based lip-reading. The research focuses on a dataset obtained through this RFID smart mask, specifically examining vowel sounds (A, E, I, O, U). Classification of the dataset through established machine learning models yields high accuracy, with the Random Forest, k-NN and SVM with RBF kernel achieving an optimal classification accuracy of 80.7% on the combined RFID dataset.