Improving Speech Inversion Through Self-Supervised Embeddings and Enhanced Tract Variables