Research map / Large language models / Language, safety and interpretability
AudioTrust: Benchmarking The Multifaceted Trustworthiness of Audio Large Language Models
Kai Li, Can Shen, Yile Liu, Jirui Han, Kelong zheng, Xuechao Zou, Lionel Z. WANG, Shun Zhang, Xingjian Du, Hanjun Luo, Yingbin Jin, Xinxin Xing and 22 more
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 privacy-enhancing prompts (ablation) | ALLMs (average) · AudioTrust privacy, direct leakage | refusal rate: 29.99 -> 63.77 (GPT-4o evaluator) | Improved |
| add privacy-enhancing prompts (ablation) | ALLMs (average) · AudioTrust privacy, inference leakage | refusal rate: 9.02 -> 12.12 | No clear effect |
| add privacy-aware prompt to Qwen2.5-Omni (ablation) | Qwen2.5-Omni · AudioTrust privacy, direct leakage | refusal rate: 0.00 -> 1.17 | No clear effect |
| closed-source vs open-source models | ALLMs · AudioTrust robustness | score: closed average generally above open-source | Improved |
| closed-source vs open-source models on voice cloning defense | ALLMs · AudioTrust authentication, VCS | defense success: open avg 45.0 vs closed avg 44.9 | Mixed |
| closed-source vs open-source on identity verification bypass | ALLMs · AudioTrust authentication, IVB | defense success: open avg 55.3 vs closed 97.2 | Improved |
| stricter speech-text prompting against voice cloning (ablation) | ALLMs · AudioTrust authentication | defense success | Improved |
| closed-source vs open-source on fairness | ALLMs · AudioTrust fairness | Gamma: GPT-4o Audio 0.926 vs Gemini-2.5 Pro 0.319 (stereotype) | Mixed |
Closest papers by meaning
- ALMGuard: Safety Shortcuts and Where to Find Them as Guardrails for Audio–Language Models (NeurIPS 2025 · similarity 0.88)
- TTA-Bench: A Comprehensive Benchmark for Evaluating Text-to-Audio Models (AAAI 2026 · similarity 0.88)
- Jailbreak-AudioBench: In-Depth Evaluation and Analysis of Jailbreak Threats for Large Audio Language Models (NeurIPS 2025 · similarity 0.88)
- AudioMarkBench: Benchmarking Robustness of Audio Watermarking (NeurIPS 2024 · similarity 0.87)
- MultiTrust: A Comprehensive Benchmark Towards Trustworthy Multimodal Large Language Models (NeurIPS 2024 · similarity 0.87)
- Audio Large Language Models Can Be Descriptive Speech Quality Evaluators (ICLR 2025 · similarity 0.87)
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.
