In this paper, we describe our pioneering work in developing speech synthesis beyond the Text-To-Speech paradigm. We introduce tangible speech synthesis as an alternate way of envisioning how artificial speech content can be produced. Tangible speech synthesis refers to the ability, for a given system, to provide some physicality and interactivity to important speech production parameters. We present MAGE, our new software platform for high-quality reactive speech synthesis, based on statistical parametric modeling and more particularly hidden Markov models. We also introduce a new HandSketch-based musical instrument. This instrument brings pen and posture based interaction on the top of MAGE, and demonstrates a first proof of concept.