Research map / Adversarial robustness and security / Attacks on text and graphs
Exploring Security Vulnerabilities in Multilingual Speech Translation Systems via Deceptive Inputs
Chang Liu, Haolin Wu, Xi Yang, Kui Zhang, Cong Wu, Weiming Zhang, Nenghai Yu, Tianwei Zhang, Qing Guo, Jie Zhang
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 more seen attack languages (EN to EN+ZH+DE+FR) (ablation) | perturbation attack on Seamless Large · Spanish unseen, eps=0.1 | ASR: 22/60 -> 55/60 | Improved |
| add more seen attack languages (EN to EN+ZH+DE+FR) (ablation) | adversarial music on Seamless Large · Spanish unseen | ASR: 4/10 -> 10/10 | Improved |
| add more seen attack languages (EN to EN+FR+DE) (ablation) | perturbation attack on Canary · Spanish unseen, eps=0.5 | ASR: 1/10 -> 6/10 | Improved |
| add Target Cycle Optimization (ablation) | perturbation attack · Seamless Large, eps 0.5/0.1/0.01 | ESIM, NSCORE | Improved |
| add Target Cycle Optimization (ablation) | adversarial music · Seamless Large | ESIM, NSCORE | Improved |
| reduce perturbation bound eps from 0.5 to 0.01 | perturbation attack · Seamless Large English | ASR: 59/60 -> 32/60 | Hurt |
| add bandpass-filtered adversarial perturbation instead of random noise (ablation) | perturbation attack · eps=0.1 EN attack | VSIM-E: 0.4172 vs 0.1452 | Improved |
| apply 8-bit quantization to adversarial audio | perturbation attack · Seamless Large eps=0.1 | ASR: 52/60 -> 12/60 | Hurt |
| apply noise addition (SNR 64 dB) to adversarial audio | perturbation attack · Seamless Large eps=0.1 | ASR: 52/60 -> 45/60 | Hurt |
| apply 12 kHz resampling to adversarial audio | adversarial music · Seamless Large | ASR: 9/10 -> 4/10 | Hurt |
| apply 6 kHz low-pass filter to adversarial audio | adversarial music · Seamless Large | ASR: 9/10 -> 8/10 | Hurt |
| change music style prompt from Techno to Classical or Orchestral (ablation) | adversarial music · English target, Seamless Large | ASR: 9/10 -> 10/10 | Mixed |
Closest papers by meaning
- AdvWave: Stealthy Adversarial Jailbreak Attack against Large Audio-Language Models (ICLR 2025 · similarity 0.87)
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models (NeurIPS 2025 · similarity 0.87)
- E2E-VGuard: Adversarial Prevention for Production LLM-based End-To-End Speech Synthesis (NeurIPS 2025 · similarity 0.86)
- Jailbreak-AudioBench: In-Depth Evaluation and Analysis of Jailbreak Threats for Large Audio Language Models (NeurIPS 2025 · similarity 0.85)
- Rethinking Audio-Visual Adversarial Vulnerability from Temporal and Modality Perspectives (ICLR 2025 · similarity 0.85)
- De-AntiFake: Rethinking the Protective Perturbations Against Voice Cloning Attacks (ICML 2025 · similarity 0.84)
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.
