Modeling dynamic vocal tracts requires accurate interpolation between target vowels for diphthong synthesis. High-fidelity Finite Element models are computationally expensive, often needing complete remeshing of the computational domain to maintain interpolation stability. In contrast, Digital Waveguide models (DWM) suffer from discretization errors due to staircased approximations of tract geometries. We model dynamic tracts using the 2D Immersed Boundary Method, which represents tract contours in the Lagrangian domain and uses a 2D Finite-Difference Time-Domain scheme to solve wave equations in the Eulerian domain. This framework enables free boundary motion within a fixed Cartesian grid, avoiding remeshing and staircasing. We synthesized three diphthongs and compared their spectral features with a 3D DWM model and recorded speech. Results show that the F1-F2 trajectories closely align with those from the 3D DWM and recorded speech, achieving correlation coefficients exceeding 0.8.