In this contribution, we will discuss a prototype that allows a group of users to design sound collaboratively in real time using a multi-touch tabletop. We make use of a machine learning method to generate a mapping from perceptual audio features to synthesis parameters. This mapping is then used for visualization and interaction. Finally, we discuss the results of a comparative evaluation study.