arXiv cs.CL· Ryo Fukuda, Atsushi Ando, Hiroki Kanagawa, Takatomo Kano, Marc Delcroix, Naohiro Tawara, Yuya Chiba·· 6 天前AI 评分24
M3-DuplexBench：面向全双工语音对话模型的多轮、多语言、多领域基准
M3-DuplexBench: A Multi-Turn, Multilingual, Multidomain Benchmark for Full-Duplex Spoken Dialogue Models
AI 导读
M3-DuplexBench 是一个面向全双工语音对话系统（FDSDS）的多轮、多语言、多领域基准，可对交互时序与回复内容进行受控评估。该基准覆盖英语和日语，横跨闲聊与多轮问答，其中英日问答数据源自同一语料库，支持跨语言的时序与内容对比。它还提供参考条件多轮设置作为可选诊断，实验揭示不同模型在交互模式、语言与领域差异以及对话上下文影响上的模型依赖性。
正文 · 原文
Abstract:Full-duplex spoken dialogue systems (FDSDSs) can listen while speaking, enabling natural conversational behaviors such as smooth turn-taking, backchannel handling, and user barge-in handling. However, controlled comparisons of FDSDSs in multi-turn interaction remain challenging, and existing benchmarks offer limited joint coverage of multiple languages, dialogue settings, interaction timing, and response content. We propose M3-DuplexBench, a multi-turn, multilingual, multidomain benchmark for controlled evaluation of FDSDSs. The benchmark jointly evaluates interaction timing and response content in English and Japanese across casual conversation and multi-turn question answering. The English and Japanese question-answering data are derived from a common source corpus, enabling cross-lingual comparisons of timing and content performance. We also provide a reference-conditioned multi-turn setting as an optional diagnostic for examining sensitivity to dialogue history and trajectory mismatch. Experiments with diverse FDSDSs reveal model-specific interaction patterns, differences across languages and domains, and model-dependent effects of dialogue context on both timing and content. The benchmark is available at this https URL.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.29125 [cs.CL] |
| (or arXiv:2607.29125v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.29125 arXiv-issued DOI via DataCite |
Submission history
From: Ryo Fukuda [view email]
[v1]
Fri, 31 Jul 2026 07:56:25 UTC (1,792 KB)
[v2]
Fri, 9 Oct 2026 10:41:08 UTC (1,829 KB)
来源：arXiv cs.CL · arxiv.org