Voiceprint authentication is a growing field of biometric authentication. However, voiceprint authentication is vulnerable to spoofing attacks and suffers significant degradation in noisy environments. Employing lip motion detection is an innovative solution to enhance security. By strategically selecting a clean channel, such as the ultrasonic frequencies generated by smartphones’ inherent audio capability, lip motion detection can be performed without disturbance from ambient noise. In this paper, we propose LV-auth, a smartphone-based multimodal biometric authentication system that integrates lip motion and voiceprint. LV-auth captures both the speech signals and the ultrasonic echo signals modulated by lip motion, enabling liveness detection while enhancing robustness. We employ a contrastive learning-based feature extractor to extract reliable and effective fused features from both modalities for authentication. Extensive experiments show that LV-auth achieves an authentication accuracy of 99.37% and an equal error rate (EER) of 0.73% on the smartphone. In particular, LV-auth demonstrates excellent robustness in noisy environments and effectively withstands spoofing attacks.