Adversarial Example Devastation and Detection on Speech Recognition System by Adding Random Noise

The automatic speech recognition (ASR) system based on deep neural network is easy to be attacked by an adversarial example due the vulnerability of network, which a hot topic in recent years. does harm ASR system, especially if common-dependent goes wrong, it will lead serious consequences. To improve robustness and security defense method against examples must proposed. Based this idea, we propose algorithm devastation detection can attack current advanced system. We choose text-dependent command-dependent as our target Generating OPT GA-based ASR. main idea input transformation examples. Different random intensities kinds noise are added devastate perturbation previously normal From experimental results, performs well. For examples, original similarity before after adding reach 99.68%, 0%, rate 94%.