Research map / Adversarial robustness and security / Attacks on text and graphs
Weighted-Sampling Audio Adversarial Example Attack
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use weighted-sampling perturbations with TVD loss instead of the C&W attack | DeepSpeech audio targeted attack · Mozilla Common Voice (first 100 test instances) | Efficiency (s): ≈3600 -> 251 | Helped |
| Use weighted-sampling perturbations with TVD loss instead of the C&W attack | DeepSpeech audio targeted attack · Mozilla Common Voice (first 100 test instances) | dB_x(δ): 38 -> 46.92 | Helped |
| Use weighted-sampling perturbations with TVD loss instead of the C&W attack | DeepSpeech audio targeted attack · Mozilla Common Voice (first 100 test instances) | Success Rate: 1 -> 1 | No clear effect |
| Use weighted-sampling perturbations with TVD loss instead of CommanderSong | DeepSpeech audio targeted attack · Mozilla Common Voice (first 100 test instances) | SNR: 17.2 -> 31.9 | Helped |
| Use weighted-sampling perturbations with TVD loss instead of EOT | DeepSpeech audio attack · Mozilla Common Voice, noise Δ=15 | Robustness: EOT-based (Δ=15): 0.07; SPT (5%): 0.4; SPT-EOT-based (75%,30): 0.30 | Mixed |
| Combine SPT with EOT | DeepSpeech audio attack · Mozilla Common Voice, noise Δ=5 | Robustness: 0.85 vs 0.74 | Helped |
| Reduce the proportion of perturbed points from 75% to 5% (ablation) | SPT audio attack · Mozilla Common Voice, noise Δ=5 | Robustness: 0.42 -> 0.71 | Helped |
| Reduce the proportion of perturbed points from 75% to 5% (ablation) | SPT audio attack · Mozilla Common Voice, noise Δ=30 | Robustness: 0.06 -> 0.27 | Helped |
| Reduce the proportion of perturbed points from 75% to 5% (ablation) | SPT audio attack · Mozilla Common Voice (first 100 test instances) | Success Rate | Hurt |
| Replace TVD metric with L∞ distance (ablation) | Weighted-sampling DeepSpeech attack · Mozilla Common Voice (first 100 test instances) | SNR: 31.9 -> 29.17 | Hurt |
| Replace TVD metric with L2 distance (ablation) | Weighted-sampling DeepSpeech attack · Mozilla Common Voice (first 100 test instances) | Success Rate: 1 -> 1 | No clear effect |
| Replace TVD metric with cosine distance (ablation) | Weighted-sampling DeepSpeech attack · Mozilla Common Voice (first 100 test instances) | Success Rate: 1 -> 0.98 | Hurt |
Closest papers by meaning
- Imperceptible, Robust, and Targeted Adversarial Examples for Automatic Speech Recognition (ICML 2019 · similarity 0.91)
- Characterizing Audio Adversarial Examples Using Temporal Dependency (ICLR 2019 · similarity 0.89)
- Defending against Adversarial Audio via Diffusion Model (ICLR 2023 · similarity 0.87)
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.84)
- Rethinking Audio-Visual Adversarial Vulnerability from Temporal and Modality Perspectives (ICLR 2025 · similarity 0.84)
- Breaking Adversarial Transferability in Fine-Tuned Speech Recognition (NeurIPS 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.