该数据集是一个大规模音频 – 文本基准数据集，作为首个专为音频大模型量身打造的多维信任评估基准，AudioTrust 专注于评估音频大语言模型（ALLMs）的多维度可信性，相关论文成果为：「AudioTrust: Benchmarking the Multifaceted Trustworthiness of Audio Large Language Models」。 数据集概括：

• 包含 4,420+ 条真实场景音频 – 文本数据，涵盖日常对话、紧急呼叫、语音助手等 18 种现实场景。
• 任务类型：音频反欺骗检测、语音指令识别、内容真实性验证、多模态语义对齐。
• 模态：音频（语音指令、欺骗性音频）、文本（指令描述、评估标签）。

Citation

@misc{audiotrust2025, title={AudioTrust: Benchmarking the Multifaceted Trustworthiness of Audio Large Language Models}, author={ Kai Li and Can Shen and Yile Liu and Jirui Han and Kelong Zheng and Xuechao Zou and Lionel Z. Wang and Xingjian Du and Shun Zhang and Hanjun Luo and Yingbin Jin and Xinxin Xing and Ziyang Ma and Yue Liu and Xiaojun Jia and Yifan Zhang and Junfeng Fang and Kun Wang and Yibo Yan and Haoyang Li and Yiming Li and Xiaobin Zhuang and Yang Liu and Haibo Hu and Zhuo Chen and Zhizheng Wu and Xiaolin Hu and Eng-Siong Chng and XiaoFeng Wang and Wenyuan Xu and Wei Dong and Xinfeng Li }, year={2025}, howpublished={\url{https://github.com/JusperLee/AudioTrust}}, }