Unvoiced Vowel Recognition using Active Bio-Acoustic Sensing for Silent Speech Interaction

In this study, we propose a silent speech recognition technique using Active Bio-Acoustics Sensing. The proposed method attaches a piezoelectric buzzer and a piezoelectric microphone to the skin surface around one ear. By sending ultrasonic swept sine signals with a frequency range of 20 kHz to 48 kHz between the piezoelectric elements, the shape of the mouth is recognized based on the frequency transfer characteristics and Support Vector Machine (SVM) classifier. We conducted experiments with seven volunteers to evaluate the recognition accuracy of a mouth forming the five Japanese vowels (‘a’, ‘i’, ‘u’, ‘e’, ‘o’) and a closed mouth. The results show an average recognition accuracy of 88.4% for the six types of mouth shapes. The proposed method enables the hands-free operation of computers while maintaining privacy and being hidden from others, even when wearing a mask or similar covering.