Enhancing Privacy Preservation with Quantum Computing for Word-Level Audio-Visual Speech Recognition

Chang Wang, Jun Du, Hang Chen, Ruoyu Wang, Chao-Han Huck Yang, Jiangjiang Zhao, Yuling Ren, Qinglong Li, Chin-Hui Lee

Abstract: In this paper, we investigate the effectiveness of using quantum machine learning for privacy protection in audiovisual speech processing. Quantum machine learning has made significant theoretical advancements and has been shown to possess natural advantages in privacy protection over conventional techniques. Here, we first apply quantum circuits to a word-level audio-visual speech recognition task. We then propose a novel metric, an inter-class intra-class similarity ratio, for measuring the privacy-protecting capabilities of quantum circuits. Finally, we conduct an in-depth analysis of the differences in privacy protection between quantum privacy methods and traditional methods, evaluating their working principles, strengths, and limitations. Experiments results on the LRW data set show that the quantum privacy-preserving approach performs well in word-level speech recognition tasks, demonstrating excellent privacy-preserving capabilities through selective retention of features.