Turn gaps average ~200ms: duplex voice models must plan replies mid-listen
smolix · x · 2026-11-07
- 跨 10 种语言的对话轮次间隔平均约 200ms(Stivers et al. 2009),而开口说一个词需约 600ms(Levinson & Torreira 2015)——人类靠在对方说话时就规划回答来跟上节奏。
- Higgs Duplex 采用同样策略:每 160ms 步是一次决策(发言、沉默、附和、让步、打断);步内固定输出 2 帧音频 + 2 帧面部隐变量,不再重复决策。
- 推理、工具与委托跨多步在后台运行,不阻塞实时对话。
More from Multimodal
- AI rotoscoping workflow today: LTX 2.5 Alpha LoRA + SAM3 + BiRefNet, layered ProRes — PurzBeats · 2026-11-07
- Custom H3 I2V/REF2V ComfyUI workflow: 20s video in 22 minutes on a 3090, controlling 5 characters — EQLMonk · 2026-11-07
- Flux Max test: mantis shrimp vs white shark prompt fails hilariously, rest looks cinematic — flowersslop · 2026-11-07
- AI video of a cat approving a transaction, made with Seedance 2.5 — prompt included — azed_ai · 2026-11-07
- Krea AI double exposure demo blends portrait and cityscape into film-like photo — azed_ai · 2026-11-07
- Fully AI-generated rapper debut: Claude, Suno, Seedance, MiniMax pipeline — Ror_Fly · 2026-11-07