Physical models of the vocal tract need a suitable voice source to generate vocalic sounds. Here we constructed and tested three types of voice sources:a self-oscillating silicone model of the vocal folds, a vibrating reed source, and asource based on an electro-mechanical loudspeaker. Each of the sources was usedto excite eight physical resonators representing the German tense vowels. The excitation signals generated by the sources were analyzed and the intelligibility of thegenerated vowel sounds was perceptually evaluated. The spectral slope of the excitation was steepest for the silicone vocal folds, least steep for the reed source, andin-between for the loudspeaker source. The intelligibility of the vowels was highest for the excitation with the silicone vocal folds, and lowest for the loudspeakersource.