📄 中文摘要
对抗样本是指通过微小的扰动使机器学习模型产生错误输出的输入数据。在自动语音识别（ASR）系统中，对抗样本的研究显示出其脆弱性和潜在的安全风险。这些对抗样本可以通过改变音频信号的特定频率或添加背景噪声来实现，导致ASR系统误识别或完全失效。这种现象不仅影响了语音识别的准确性，还可能在安全和隐私方面带来严重后果。研究者们正在探索如何增强ASR系统的鲁棒性，以抵御这些对抗攻击，并确保语音识别技术的可靠性和安全性。
你听到了吗？针对自动语音识别的对抗样本
📄 English Summary
Did you hear that? Adversarial Examples Against Automatic Speech Recognition
Adversarial examples refer to inputs that are perturbed in subtle ways to cause machine learning models to produce incorrect outputs. Research on adversarial examples in automatic speech recognition (ASR) systems reveals their vulnerabilities and potential security risks. These adversarial samples can be created by altering specific frequencies of audio signals or adding background noise, leading to misrecognition or complete failure of ASR systems. This phenomenon not only affects the accuracy of speech recognition but also poses serious implications for security and privacy. Researchers are exploring methods to enhance the robustness of ASR systems against such adversarial attacks, ensuring the reliability and security of speech recognition technologies.