| --- |
| license: mit |
| --- |
| ### **Dataset Name: AudioInject-Robust** |
| |
| **Version:** 1.0 |
| |
| **Release Status:** ⏳ **Coming Soon** (The data is currently being prepared for public release and will be uploaded shortly. Stay tuned!) |
| |
| ### **Dataset Description** |
| |
| `AudioInject-Robust` is a benchmark dataset specifically designed to evaluate the robustness of Large Audio-Language Models (LALMs) against malicious audio injection attacks. |
| As LALMs are increasingly deployed in real-world applications, they face the risk of being manipulated by malicious audio, which can lead to outputs that deviate from user expectations, generate biased responses, or make unfair judgments. This dataset aims to systematically evaluate the vulnerabilities and resilience of models across multiple attack scenarios, supporting the development of safer and more reliable LALMs. |
