Scientists have used ultrasound to examine speech articulation for over 50 years. This benign, noninvasive technique has become increasingly popular as image quality and analysis methods have improved. Most work involves ultrasound imaging of the tongue (UIT) and can be compared with such alternatives as point parameterization (e.g., electromagnetic articulometry, EMA) and real-time magnetic resonance imaging (rtMRI). EMA provides excellent spatial and temporal resolution but sparse coverage. UIT typically has a slower frame rate than EMA but faster than rtMRI. Both UIT and rtMRI provide substantially greater coverage of the tongue surface than EMA. For consistent comparison of tongue shapes, UIT requires tracking the position and orientation of the probe relative to the head to accommodate effects of jaw displacement; thus, co-collection of EMA and UIT are increasingly common, providing the complementary advantages of both. Biofeedback provided by UIT has been used to support remediation of misarticulations in a first language, decreasing a second language accent, and improving singing technique. New machine learning approaches to discretizing tongue shapes from video are enhancing automated analysis. Modern ultrasound has also been applied to imaging laryngeal devoicing gestures, and the final part of the presentation will show progress in quantifying such gestures.
Whalen et al. (Wed,) studied this question.