The holy grail for violin makers is to find correlations between construction parameters and sound qualities. This is challenging for two main reasons: it is difficult to build violins reliably enough to ensure that the change in the sound is indeed a result of the change of construction parameters; when listening to the violins being played, differences seem to be smoothed out by the players who adapt very quickly. Therefore, while players had so far been preferred in our experiments to maximize the ecological validity and take into account the complexity of the interaction between the player and the instrument, we have decided to test whether other methods, which reduce the influence of the player but are quite artificial, may be useful to explore the influence of some construction parameters on the tone. In the context of two sets of violins built with controlled thickness variations of their plates, we will compare the results of listening tests based on real recordings with a player and with a bowing machine as well as synthesized recordings (from the convolution of an excerpt recorded with piezo sensors and radiation measurements in an anechoic chamber) and discuss them in the light of audio descriptors on the recordings and vibroacoustical measurements on the violins.