This work aims to explore the use of a new gesture-based interaction built on automatic recognition of Soundpainting structured gestural language. In the proposed approach, a composer (called Soundpainter) performs Soundpainting gestures facing a Kinect sensor (Microsoft). Then, a gesture recognition system captures gestures that are sent to a sound generator software. The proposed method was used to stage an artistic show in which a Soundpainter had to improvise with 6 different gestures to generate a musical composition from different sounds in real time. The accuracy of the gesture recognition system was evaluated as well as Soundpainter's user experience. In addition, a user evaluation study for using our proposed system in a learning context was also conducted. Current results open up perspectives for the design of new artistic expressions based on the use of automatic gestural recognition supported by Soundpainting language.