Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Music: Real world Audio Adversary against Wake-word Detection System
Juncheng Li, Shuhui Qu, Xinjian Li, Joseph Szurley, J. Zico Kolter, Florian Metze
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Add adversarial music | Emulated wake-word model · Real-world utterances, digital | F1: 0.955 -> 0.117 | Hurt |
| Add adversarial music | Emulated wake-word model · Real-world utterances, physical | F1: 0.934 -> 0.110 | Hurt |
| Add adversarial music | Amazon Alexa · Real-world utterances, physical | F1: 0.925 -> 0.110 | Hurt |
| Start adversarial music 1 second after the wake-word instead of overlapping it (ablation) | Adversarial Music on Alexa · Alexa physical test, φ = 0°, dₜ=4.2ft | wake-word responses: 0/10 -> 10/10 | Hurt |
| Start adversarial music 1 second before the wake-word (ablation) | Adversarial Music on Alexa · Alexa physical test, φ = 0°, dₜ=4.2ft | wake-word responses: 10/10 -> 0/10 | Improved |
| Play adversarial music continuously (ablation) | Adversarial Music on Alexa · Alexa physical test, φ = 0°, dₜ=4.2ft | wake-word responses: 10/10 -> 0/10 | Improved |
| Replace adversarial music with random music at the same volume | Alexa wake-word detection · Alexa physical test, φ = 0°, dₐ=7.7ft, dₜ=4.2ft, 70 dBA | wake-word responses: 2/10 -> 10/10 | Hurt |
| Replace adversarial music with random single notes at the same volume | Alexa wake-word detection · Alexa physical test, φ = 0°, dₐ=7.7ft, dₜ=4.2ft, 70 dBA | wake-word responses: 2/10 -> 9/10 | Hurt |
| Replace adversarial music with real guitar music at the same volume | Alexa wake-word detection · Alexa physical test, φ = 0°, dₐ=7.7ft, dₜ=4.2ft, 70 dBA | wake-word responses: 2/10 -> 10/10 | Hurt |
| Remove room impulse response transformations from attack optimization (ablation) | Adversarial Music · Alexa physical, over the air | attack effectiveness | Hurt |
| Increase adversarial music volume from 40 dBA to 70 dBA | Adversarial Music on Alexa · Alexa physical, varied distances and angles | attack success | Improved |
| Increase attacker-to-Alexa distance | Adversarial Music on Alexa · Alexa physical, varied distances | attack effectiveness | Hurt |
Closest papers by meaning
- Hearing Without Noticing? Attention-Aware Stealthy Black-Box Adversarial Audio Attacks (ICML 2026 · similarity 0.87)
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.86)
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models (NeurIPS 2025 · similarity 0.83)
- E2E-VGuard: Adversarial Prevention for Production LLM-based End-To-End Speech Synthesis (NeurIPS 2025 · similarity 0.82)
- Rethinking Audio-Visual Adversarial Vulnerability from Temporal and Modality Perspectives (ICLR 2025 · similarity 0.82)
- Adaptive Compression and Targeted Perturbation: A Unified Framework for Generalized Audio Deepfake Detection (NeurIPS 2026 · similarity 0.81)
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.
