VocalNet/CS3-Bench

CS3-Bench是一个用于评估和增强多模态大型语言模型在语音到语音交互系统中语言对齐能力的代码转换语音到语音基准数据集，特别针对普通话-英语代码转换。该数据集包括三种配置：知识、知识_en和开放性问题，每种配置都具有特定的特征，如问题ID、音频、问题和答案（中英两种语言）。它还包含了不同模型在知识密集型问答和开放式对话中的评估结果。CS3-Bench is a speech-to-speech code-switching benchmark dataset designed to evaluate and enhance the language alignment capabilities of multimodal large language models in speech-to-speech interaction systems, particularly for Mandarin-English code-switching. The dataset includes three configurations: knowledge, knowledge_en, and open_ended, each with specific features like Qid, audio, Question, and Answer in both Chinese and English. It also contains evaluation results for different models on both knowledge-intensive question answering and open-ended conversations.