Physical replicas of the vocal tract are often produced by 3D-printing. A proven method to obtain the acoustic transfer function of these models is based on the principle of reciprocity, where an external sound source is used and the acoustic response is measured inside the model at the glottal end. However, earlier investigations showed that for some model geometries the measured transfer functions deviated from the expected resonance pattern due to spurious poles and zeros. These artifacts are probably caused by sound transmission through the model walls, or by whole-body vibrations of the replicas. In this empirical study, two methods to reduce these artifacts were compared, namely wrapping the models with sound-absorbing fabric and embedding the models in sand. Axisymmetric 3D-printed tubes for ten different vowels were analyzed. The results showed that both methods reduce artifacts and at the same time improve the repeatability of the measurements.