LiveKit: voice agents feel human around 600ms latency, not leaderboard scores
AI Engineer · youtube · 2026-10-11
Jesse Hall of LiveKit (AI Engineer World's Fair 2026) explains building voice agents that feel human:
- The leaderboard trap: scores measure one model, but you ship a pipeline — errors compound across STT → LLM → TTS
- Measured vs perceived latency: a fast dashboard can still feel robotic to callers
- The window that matters: 1.5s end-to-end still works, 600ms starts to feel human; treat latency as a budget
- Audio-based turn detection: 4.5% of users cut off vs 9.9% for the best alternative, plus buying time during tool calls like a person does
- Includes a hotel-receptionist demo handling interruptions, an open-source benchmark to run on your own agent, a voice-tuned Gemma preview on LiveKit Inference, and a checklist
More from coding & agent
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23
- New data marketplace lets vendors list repos; coding agents access via MCP server — testingcatalog · 2026-10-23
- Tie code approvals to content hashes so agent 'one more small fix' voids review — Ambitious_Answer6857 · 2026-10-23