An experimental set-up and human vocal folds replica able to produce self sustained oscillations is presented. The aim of the set-up is to assess the relevance and the accuracy of theoretical vocal folds models. The applied reduced mechanical models are a variation of the classical two-mass model. The airflow is described as a laminar flow with flow separation. The influence of a downstream resonator is taken into account. The oscillation pressure threshold and fundamental frequency are predicted by applying a linear stability analysis to the mechanical models. The measured frequency response of the mechanical replica together with the initial (rest) area allows to determine the model parameters (spring stiffness, damping, geometry, masses). Validation of theoretical model predictions to experimental data shows the relevance of low order models in gaining a qualitative understanding of phonation. However quantitative discrepancies remain large due to an inaccurate estimation of the model parameters and the crudeness in either flow or mechanical model description. As an illustration it is shown that significant improvements can be made by accounting for viscous flow effects.