Silent Speech Recognition based on Short-Range Millimeter-Wave Sensing

A new silent-speech recognition technique based on a compact custom-designed 120 GHz interferometric radar system is proposed. The millimeter-wave 120 GHz makes it possible to precisely track the fine displacement of the mouth movement, which is strictly related to the speech activities. To overcome the phase ambiguity in millimeter-wave non-linear phase modulation, a novel algorithm based on trigonometric transform is proposed to linearly reconstruct the phase information of the speech. The proposed technique carries the merits of contact-free, high penetrability and high precision. Experiments have been performed in the office environment and the results show that the proposed technique performs well in sensing the precise silent-speech commands for both words and sentences. The specific patterns existing in each command provide foundation for silent-speech recognition.