Weighted-Sampling Audio Adversarial Example Attack
Authors: Xiaolei Liu, Kun Wan, Yufei Ding, Xiaosong Zhang, Qingxin Zhu4908-4915
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that our method is the first in the field to generate audio adversarial examples with low noise and high audio robustness at the minute timeconsuming level 1. Finally, our experiments show that our method can generate more robust audio adversarial examples in a short period of 4 to 5 minutes. This is a substantial improvement compared to the state-of-the-art methods. |
| Researcher Affiliation | Academia | 1University of Electronic Science and Technology of China, 2University of California, Santa Barbara |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper provides a link to listen to audio adversarial examples (https://sites.google.com/view/audio-adversarial-examples/) but does not provide concrete access to the source code for the methodology described in the paper. |
| Open Datasets | Yes | Dataset. Mozilla Common Voice dataset2 (MCVD): MCVD is an open and publicly available dataset of voices that everyone can use to train speech-enabled applications. It consists of voice samples require at least 70GB of free disk space. 2https://voice.mozilla.org/en/datasets |
| Dataset Splits | No | The paper states 'We follow the convention in the field and use the first 100 test instances of this dataset to generate audio adversarial examples' but does not specify training/validation/test splits for the overall experimental setup. |
| Hardware Specification | Yes | Environment. All experiments are carried out on an Ubuntu Server (16.04.1) with an Intel(R) Xeon(R) CPU E5-2603 @ 1.70GHz, 16G Memory and GTX 1080 Ti GPU. |
| Software Dependencies | No | The paper mentions 'Ubuntu Server (16.04.1)' but does not provide specific version numbers for other key software components, libraries, or frameworks used in the experiments. |
| Experiment Setup | Yes | For reproducibility, here we give the hyperparameters used in our experiments. The max iteration is set to be 500... In SPT, the proportion of perturbed points is 75%. In WPT, we set the weights of key points to be 1.2, the learning rate begins with 100 and β is set to be 0.8. The hyperparameters c and γ are 0.001 and 10. |