Flow-visualization with smoke particles illuminated by a laser sheet, performed with a high-speed camera, was used to obtain the flow patterns in two models of the human larynx, both scaled by a factor of 7.5. The first model was static, while the second was a driven dynamic mechanical model, which simulated the motion of the vocal folds during phonation. For the dynamic model, the glottal motion frequency was of the order of the average fundamental frequency of males. It was found that the vortical flow structures were shed at different locations within and downstream the glottis, depending on the transglottal pressure, geometrical configuration and glottal motion frequency. The vortex shedding frequency, which was much higher than the motion frequency of the vocal folds, was measured optically based on the high-speed visualization images, and compared to the frequency peaks in the sound spectra, determined by microphone measurements. Generally, it was observed that the vortex sheddeing frequency corresponds to the main peak in the sound spectrum. The intensity of noise, however, was found to be dependnet of the location of vortical structures relative to the glottal walls, indicating the presence of quadrupole- or dipole-type sound sources, due to ordered flow structures.