An Ultra-Sensitive Humidity Sensor for Silent Speech Intelligent Recognition

Abstract Globally, millions of individuals lost ability to speak due to various medical conditions, rendering them unable to engage in voice-based communication or speech recognition. Silent speech recognition represents a groundbreaking technology that aims to bridge this communication gap, by providing an auxiliary means of interaction for those who cannot speak. However, conventional contact-based silent speech recognition systems often suffer from data inaccuracies caused by positional shifts resulting from muscle contractions and relaxations during articulation. Furthermore, prolonged use of these systems can lead to discomfort and inconvenience for users. To address these challenges, we propose a non-contact silent speech recognition technology that utilizes the unique detection capabilities of humidity sensors. Our research employs oxygen plasma reactive ion etching to fabricate hydrophilic nanoforest materials on an 8-inch wafer, achieving an exceptionally high specific surface area. The three-dimensional upright structure of these materials facilitates rapid and deep transport of water molecules, enabling a humidity sensor with remarkable sensitivity (0.913 pF/% RH). Furthermore, this sensor is incorporated into a non-contact silent speech recognition system that employs wireless transmission modules and convolutional neural network algorithms to achieve ultrafast response (0.57 s) and high-precision intelligent recognition of silent speech, including words, phrases, and short sentences, with an accuracy rate of 98.51%. This system holds promise for applications in medical monitoring, virtual reality, and specialized security scenarios, offering innovative solutions for silent command recognition and enhanced communication across various fields.