This paper proposes a novel speech acquisition method to directly sense the sound waves travelling through the air rather than leveraging the sound-induced vibration. The rationale is that the piezoelectric film can cause changes to the incident mmWave signals due to the piezoelectric effect when sound waves hit the film. Moreover, we find that the mmWave signals can penetrate common soundproofing materials, which pave the way for through-obstacle speech recovery. Based on this, we propose a new mechanism to decode the phase-demodulated audio signals from the reflected mmWave signals. To combat the penetrating and propagation loss of mmWave signals, we further developed a speech enhancement method based on the deep neural network. The network can filter the raw recovered speech and reconstruct clean speech with higher speech quality and intelligibility. We evaluate the proposed system on a public dataset. The results indicate that the system can recover high-quality and intelligible speech from a distance of 5m in the case of occlusion.