The resonances of the vocal tract determine the different vowels of speech. But beyond that, they are a key factor in voice quality and singing techniques. Very simple and efficient acoustic models are extensively used to simulate vocal tract acoustics in the frequency range for which the plane wave assumption is valid (up to 4-5 kHz). However, outside of this range, more complex models must be used. A trade-off must be found between the accuracy of the description of the vocal tract geometry and the computational cost. Thus, as an example, 3D finite elements allow one to take into account all the details of the anatomy but require a lot of computational resources. On the other hand, the state-of-the-art multimodal methods can have a much lower computational cost, but require some simplification of the anatomy. The present work proposes to combine the advantages of both methods to simulate vocal tract acoustics with a reasonable computational cost and an accurate anatomical description. For this purpose, the vocal tract shape is sliced in successive planes, allowing one to simulate the multimodal propagation along a centerline connecting the frames. On each frame, the local transverse modes are computed with 2D finite elements, which allows one to take into account accurately the cross-sectional shapes.