voice-activity-detection

这篇论文提出了一套自动化流程，用于从分离声道录音的自发对话中提取对话轮次与反馈语（backchannels），在99段丹麦语对话上评估得到约0.62的F1分数，可作为半自动标注流程中的一致性初步标注。

FD-VAD proposes an ASR-free, streaming semantic endpointer that maps causal audio windows directly to Continue/Stop decisions for full-duplex voice agents, combining a frozen speech encoder with a parameter-efficiently adapted language model and achieving state-of-the-art EOT recall on TurnBench.

A detailed explainer on the five-layer architecture of AI voice agents, including speech-to-text, LLM, text-to-speech, orchestrator, and telephony, all operating under a 500ms latency constraint to maintain natural conversation flow.