FD-VAD：面向流式全双工语音的语义端点检测
原标题：FD-VAD: Semantic Endpoint Detection for Streaming Full-Duplex Speech
AI 摘要
论文提出 FD-VAD，一种无需 ASR 的流式语义端点检测方法，用于全双工语音交互。它将语义端点检测建模为因果音频-语言推理任务，把有界因果音频窗口直接映射为 Continue/Stop 决策，结合冻结语音编码器、轻量模态适配器和参数高效微调的语言模型，并采用 last-chunk 训练目标。方法还引入置信度门控端点提交与边界聚焦难负样本采样，在域内和对话评测中优于强流式与非流式语义轮次分类器，并在 TurnBench 开发集零样本设置下取得最高 EOT 召回。
正文节选
FD-VAD: SEMANTIC ENDPOINT DETECTION FOR STREAMING FULL-DUPLEX SPEECH Abstract Natural turn-taking in full-duplex voice interaction requires determining from partial speech whether a pause reflects hesitation or a completed conversational intent. Acoustic voice activity detection lacks this semantic information, while cascaded ASR-based endpointing introduces transcription dependence and additional processing stages. We formulate semantic endpoint detection as a causal audio-language reasoning ta