A real-time, automated tongue tracking method for ultrasound biofeedback in speech therapy

Abstract

Midsagittal ultrasound images of the tongue surface are currently used as feedback on tongue motion for speech therapy, but can be difficult to interpret. Ultrasound biofeedback would benefit from a simplified, immediate representation of tongue motion. Tongue tracking in ultrasound is complicated by reflections from structures near the tongue surface, complex tongue shapes, and rapid, extreme tongue movements. Available programs that trace contours of the tongue surface have not been implemented in real time and can require extensive user input. We offer a tongue tracking method that can operate in real time and requires minimal user input. This method applies a low-pass filter to the B-mode image and then maps the tongue surface based on local brightness maxima within search regions estimated by 2nd-order Taylor series expansions in space and time. Required user input includes a graphically selected calibration point near the tongue surface and relative brightness thresholds for anterior and posterior ends of the tracked surface. Preliminary results show that this method can capture complex, rapid movements of tongue parts during production of American English /r/. Efficient tongue tracking enables real-time analysis of tongue dynamics, offering a valuable tool for speech therapy.