Improved Acoustic-to-Articulatory Inversion Using Representations from Pretrained Self-Supervised Learning Models