Hyeongjun Choi, Young Eun Kwon, Ji Won Yoon (Korea University)
This paper presents mmProcess, a novel phasebased approach for speech reconstruction using millimeterwave (mmWave) technology, offering an alternative to existing Doppler-based and deep learning-dependent methods. By leveraging the phase variations in mmWave signals, mmProcess enables precise detection of fine vibrations caused by sound, facilitating accurate speech reconstruction without the need for large training datasets, prior knowledge, or complex neural networks. This eliminates the limitations of deep learning approaches, such as degraded performance with unseen languages and the significant time and cost required for system development. mmProcess combines advanced signal processing techniques, including range processing, phase unwrapping, and noise filtering, to transform raw mmWave radar data into high-fidelity speech signals. Experimental evaluations validate the effectiveness of the method, demonstrating its capability to operate in challenging scenarios while maintaining adaptability and cost efficiency.