Deep-Learning-Based Methods for Automatic Articulator and Levator Veli Palatini Segmentation and Motion Quantification in Magnetic Resonance Images of the Vocal Tract