A lightweight end-to-end anti-spoofing voice model based on WavLM

In the rapidly evolving domain of voice-assisted technologies, ensuring the authenticity and security of voice interactions is paramount, especially in scenarios involving identity verification. This paper introduces a novel, lightweight anti-spoofing voice detection model that integrates the WavLM-Base for feature extraction with a straightforward, resource-efficient classification framework. This model effectively discriminates between genuine and spoofed audio inputs, emphasizing the utility of self-supervised learning models in enhancing security measures within Automatic Speaker Verification systems without necessitating extensive computational resources. The approach leverages innovative techniques to address the growing challenges in voice security, contributing significantly to the robustness and reliability of voice-based systems. We achieve an EER of 0.45019LA dataset, demonstrating the competitive performance of our method.