A research study by Professor Ren Yanzhen's team from the School of Cyber Science and Engineering at Wuhan University (WHU) has been accepted for presentation at the 46th IEEE Symposium on Security and Privacy (IEEE S&P 2025). IEEE S&P 2025, a premier international conference in information security, will be held in May in San Francisco, the United States.  

The paper, Lombard-VLD: Voice Liveness Detection based on Human Auditory Feedback, addresses the vulnerability of automatic speaker verification (ASV) systems to voice spoofing attacks. The study proposes an innovative voice liveness detection (VLD) method that uses the Lombard effect – an involuntary speech adjustment mechanism where individuals naturally modify their speech patterns, such as increasing volume or altering pitch – in noisy environments to enhance intelligibility.

Lombard-VLD: The extracting feature differentiates Lombard from normal speech for liveness detection.

The research team developed the Lombard-VLD method using a SE-ResBlock differential network to detect acoustic differences between genuine and spoofed speech under Lombard effect stimulation. Results demonstrated that the method achieves error rates of 0 percent and 0.24 percent on two benchmark voice datasets, outperforming existing technologies. Additionally, it exhibits strong robustness and generalization capabilities across varying environments and unseen speakers.  

This study offers a novel solution for voice liveness detection, providing a crucial safeguard against deepfake fraud amid the rapid advancement of AI-generated content.