StepEval-Audio-360

该数据集是一个为评估多模态语音交互而设计的新基准，它重点关注用户体验和安全。该基准涵盖了诸如语言熟练度、情感智力、逻辑推理和创造力等评估维度，并采用了真实用户录音和公开语料库，同时对音频质量和语义标注进行了严格把控。这项任务的目的是对多模态语音交互进行综合评估。

This dataset is a novel benchmark designed for evaluating multimodal speech interactions, with a core focus on user experience and safety. The benchmark covers evaluation dimensions such as language proficiency, emotional intelligence, logical reasoning and creativity. It adopts real user audio recordings and public corpora, while imposing strict controls over audio quality and semantic annotation. The purpose of this task is to conduct comprehensive evaluations of multimodal speech interactions.