An Application Of The Subspace Approach On Radar Speech Enhancement

A new kind of 94 GHz bio-radar was developed in our lab for non-acoustic sensor speech detection. Because of the excellent performance of the 94 GHz radar, this new equipment provides a novel speech Acquisition method. However, the quality of the speech detected by 94 GHz radar was corrupted by some new kinds of noises, such as electromagnetic-harmonic-noise, radar-electro-circuit noise, and the environmental noise. Therefore, the main topic of this current study is to increase the intelligibility and the quality of the millimeter-wave-radar-speech, this study proposed a subspace approach. The subspace-speech-enhancement-algorithm decomposes the bio-radar signal vector space to two subspaces, one is signal-with-noise-subspace, the other one is noise-subspace, so the algorithm has ability to reduce the unnatural structure of residual noise. The results of the simulation, as well as speech quality evaluation confirm that the comprehensibility of the enhanced speech detected by radar is much better than the results by using other traditional speech enhancement algorithm, suggest that the proposed subspace approach has good ability to reduce the electromagnetic-noise from 94 GHz millimeter wave radar speech.