The acoustic propagation in the human vocal tract using the Transmission Line Matrix method (TLM) has been simulated. Tri-dimensional meshes are built from segmented magnetic resonance images. Initially, simple models of one and two tubes were constructed to then simulate the same signal in a human vocal tract model whose shape represents the production of the open vowel /a/, simulating at the same time a human glottal signal. For the vocal tract model, the voice signals generated by the simulations were compared with voice samples from the same subject from whom the mesh was extracted. A compensation factor for the topology-dependent numerical dispersion for improving formant determination is proposed. It is concluded that the proposed compensation factor improves the effectiveness and accuracy of the TLM method.