UCSC-VLAA/PARADE_audio

AHELM（音频语言模型全面评估）数据集是一个用于评估音频语言模型在音频感知、知识、推理、情感检测、偏见、公平性、多语言性、鲁棒性、毒性和安全性等10个关键方面的性能的基准。该数据集包含了多个子数据集，其中包括PARADE和CoRe-Bench，分别用于评估模型避免刻板印象和对话音频推理的能力。AHELM (A Holistic Evaluation of Audio-Language Models) dataset is a benchmark designed to assess the performance of audio-language models across 10 key aspects: audio perception, knowledge, reasoning, emotion detection, bias, fairness, multilinguality, robustness, toxicity, and safety. The dataset consists of multiple sub-datasets, including PARADE and CoRe-Bench, which evaluate the models ability to avoid stereotypes and reasoning over conversational audio through inferential multi-turn question answering, respectively.