Research map / Adversarial robustness and security / Attacks on text and graphs
Characterizing Audio Adversarial Examples Using Temporal Dependency
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 temporal inconsistency measured by WER to detect adversarial audio | DeepSpeech ASR · Common Voice | AUC: 0.712 -> 0.936 | Helped |
| Use temporal inconsistency measured by WER to detect adversarial audio | DeepSpeech ASR · LIBRIS | AUC: 0.645 -> 0.930 | Helped |
| Apply Quantization-256 | Speech commands classifier under GA attack · Speech Commands | average attack success rate: 84% -> 2.1% | Helped |
| Apply Quantization-256 | Speech commands classifier under GA attack · Speech Commands benign audio | accuracy: 89.2% -> 89.0% | No clear effect |
| Apply input transformations such as quantization, downsampling, or smoothing | DeepSpeech under adaptive Opt attack · LibriSpeech | attack success rate: 92%; 98%; 98% | Hurt |
| Use temporal-dependency detection against adaptive segment attacks | DeepSpeech ASR · LibriSpeech and Common Voice | attack success rate: 2%; 2% | Helped |
| Apply frame-level autoencoder transformation | DeepSpeech on benign Common Voice speech · Common Voice, language model | WER: 27.5 -> 57.6 | Hurt |
| Apply frame-level autoencoder transformation | DeepSpeech on benign LibriSpeech speech · LibriSpeech, language model | WER: 12.4 -> 30.0 | Hurt |
| Use temporal inconsistency measured by CER instead of WER (ablation) | TD detector · LIBRIS | AUC: 0.930 -> 0.933 | No clear effect |
| Use temporal inconsistency measured by LCP ratio instead of WER (ablation) | TD detector · Common Voice | AUC: 0.936 -> 0.859 | Hurt |
| Set the inspected prefix fraction to k=4/5 (ablation) | TD detector using CER · Common Voice | AUC: 0.933 -> 0.969 | Helped |
| Use adaptive combination attack against temporal-dependency detection | TD detector with random attacker and detector sections · LibriSpeech | attack success rate: 0.873 | Mixed |
Closest papers by meaning
- Imperceptible, Robust, and Targeted Adversarial Examples for Automatic Speech Recognition (ICML 2019 · similarity 0.91)
- Rethinking Audio-Visual Adversarial Vulnerability from Temporal and Modality Perspectives (ICLR 2025 · similarity 0.89)
- Weighted-Sampling Audio Adversarial Example Attack (AAAI 2020 · similarity 0.89)
- Breaking Adversarial Transferability in Fine-Tuned Speech Recognition (NeurIPS 2026 · similarity 0.85)
- DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain (NeurIPS 2024 · similarity 0.85)
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.85)
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.