Anti-Spoofing Evaluation Benchmark and Layered Defense Strategy for In-Cabin Voiceprint Payment under Voice Cloning Attacks

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

With the proliferation of voice interaction in intelligent cabins, voiceprint payment is becoming an important payment method in vehicular scenarios. However, deep learning-driven voice cloning technology has significantly lowered the barrier for generating forged speech, posing a serious threat to voiceprint authentication. The enclosed cabin environment, multi-microphone array layout, and real-time requirements make traditional anti-spoofing solutions difficult to migrate directly. This paper systematically analyzes the technological evolution of voice cloning attacks and the in-cabin threat model, constructs a multi-dimensional anti-spoofing evaluation benchmark covering detection performance, user experience, system overhead, and robustness, and proposes a four-layer defense strategy comprising input-layer liveness detection, feature-layer multimodal fusion, decision-layer risk grading, and application-layer transaction control. Experiments show that the proposed framework can reduce the attack success rate to 0.9%, while maintaining a false rejection rate of 0.7% for legitimate users and an authentication latency of 847 ms, providing a systematic reference for the security evaluation and protection of in-cabin voiceprint payment.