Real-Time Speech Recognition Using High-Frequency Micro- 2Doppler Radar

This study presents a novel approach to remote speech recognition using a millimeter- 7wave micro-Doppler radar system operating at 94 GHz. The proposed method uses high-frequency 8radar to detect subtle speech-related vibrations, enabling speech recognition that is both non-contact 9and privacy-preserving. Tests with actual human speech followed initial experiments in which a 10piezoelectric crystal was used to simulate vocal cord vibrations. Radar returns were processed using 11state-of-the-art signal processing techniques, including short-time Fourier transform (STFT), to 12generate spectrograms and reconstruct speech signals. The system demonstrated high accuracy in 13speech reconstruction, with a strong correlation between the radar-reconstructed audio and the 14original speech signals. Cross-correlation analysis quantitatively confirmed the similarity between 15the reconstructed and original audio. These results validate the system's effectiveness in detecting 16and characterizing speech-related vibrations without direct audio recording. The findings support 17this innovative approach, with significant implications for applications in security, surveillance, and 18assistive technologies where privacy-preserving solutions are essential. Future research will focus 19on diverse real-world scenarios and further integration of advanced signal processing and machine 20learning techniques to enhance accuracy and robustness.