In this work, we develop a 3D subject-specific biomechanical model of the oropharynx in order to investigate and simulate speech production. Our muscle-activated model is generated based on the subject-specific anatomy captured from dynamic volumetric cine-MRI data. Our model includes an air-tight deformable airway that enables speech synthesis. We simulate our model based on actual tissue motion tracked from the tongue during speech production, which we extract from the tagged-MRI data. We quantitatively validate our model on MRI data achieving an average target point tracking error of 1.15mm ± 0.632, and an acoustic formant frequency estimation error of 6.01% ± 4.92%.