语音对话系统必须在用户话语进行中实时区分打断、backchannel 和语轮完成。既有模块化方法在 utterance 或固定 chunk 级优化轮转状态，与连续的轮转估计不匹配，且常依赖辅助 ASR 模型，响应慢、系统复杂。
🔬 方法
基于预训练 Voxtral Realtime 模型，引入与 ASR 头并行的 frame-synchronous 轮转状态头（delayed-stream modeling），在共享流式表征上联合预测 ASR token 与帧级细粒度轮转状态。
章节结构（全文标题提取）： 1 Introduction 2 Method 3 Experimental setup 4 Results and analysis 5 Conclusion 2.1 Dual-Head Modeling · 2.2 Turn-Taking State Token Design · 2.3 ASR-Anchored Turn State Supervision · 3.1 Data Preparation · 3.2 Implementation Details · 3.3 Latency Metric · 4.1 Main results · 4.2 Effect of streaming delay τ \tau for Turn-taking
📊 结果
中英双语 Easy-Turn 测试集上实现准确的轮转检测并保持低延迟。
💡 与研究方向关联
全双工语音交互中 turn-taking 是核心难题：系统必须在对方话语进行中判断其是否完成、要不要打断，或只是 backchannel（'嗯嗯'类无意义反馈）。这与本用户 duplex 方向直接相关——DuplexOmni 用 480ms 时间片、DuplexWorld 用 200ms tick 建模用户侧打断/backchannel 决策。X2-Turn 把 turn state 预测下推到帧级并与 ASR 共享表征，是模块化的一个关键解耦。
📝 原文摘要
▶ 原文摘要 Abstract
Accurate and responsive turn-taking is essential for spoken dialogue systems, which must distinguish in real time between user interruptions, backchannels that should be ignored, and the completion of an utterance. Prior modular approaches typically optimize turn state prediction at the utterance or fixed-chunk level, creating a mismatch with the continuous turn state estimate, and often depend on an auxiliary ASR model, which limits responsiveness and increases overall system complexity. Therefore, we present X2-Turn, a frame-synchronous turn state prediction method via delayed-stream modeling. Specifically, building on the pretrained Voxtral Realtime model, we introduce a frame-synchronous turn state head that operates in parallel with the ASR head on shared streaming representations, jointly predicting ASR tokens and fine-grained turn states at the frame level. We evaluate our method on the bilingual Chinese-English Easy-Turn test sets, and the results demonstrate its effectiveness in achieving accurate turn-taking detection while maintaining low latency.