Speech sensing is presented as a novel application of ultrawideband (UWB) technology, possibly improving synthetic speech and speech pathology and allowing silent speech recognition. A finite-difference time-domain (FDTD) model for the human vocal tract is presented and compared to monostatic measurements, indicating that vocal tract dynamics involving the lips and tongue can be identified. A proof-of-concept speech recognition experiment is also presented, showing that even with a very simple algorithm, high match rate is achievable.