SFC-Bench
收藏资源简介:
SFC-Bench是首个大规模语音功能调用数据集,由上海交通大学与阿里巴巴集团联合构建。该数据集基于传统SLU基准(如ATIS、SNIPS、FSC、SLURP)精心筛选并扩展出300个口语功能,通过多智能体系统自动合成涵盖单轮、多轮及多意图等复杂度的查询与标签。其旨在为大型语言模型和大型音频语言模型提供标准化评估框架,突破传统闭集SLU在开放域语义提取中的模糊性局限,推动语音交互向结构化、可泛化的功能调用范式演进。
SFC-Bench is the first large-scale spoken function calling dataset, jointly constructed by Shanghai Jiao Tong University and Alibaba Group. Based on traditional Spoken Language Understanding (SLU) benchmarks including ATIS, SNIPS, FSC and SLURP, this dataset has been carefully screened and expanded to encompass 300 spoken functions. It automatically synthesizes queries and labels with complexities covering single-turn, multi-turn and multi-intent scenarios via a multi-agent system. This dataset aims to provide a standardized evaluation framework for large language models (LLMs) and large audio language models, breaking through the ambiguity limitations of traditional closed-set SLU in open-domain semantic extraction, and advancing the evolution of spoken language interaction towards a structured and generalizable function calling paradigm.




