🎙️ Voice AI only feels natural when conversation keeps pace with speech.

Here’s how we rebuilt our...

OpenAI Developers(@OpenAIDevs)

OpenAI Developers(@OpenAIDevs)2026年5月5日

🎙️ Voice AI only feels natural when conversation keeps pace with speech. Here’s how we rebuilt our...

7.8Score

TL;DR · AI 摘要

OpenAI 重构 WebRTC 栈，采用轻量中继与有状态转码器，显著降低语音 AI 实时延迟，支撑 ChatGPT 语音与 Realtime API 的自然对话体验。

核心要点

语音 AI 的自然感核心在于端到端延迟匹配人类语速节奏
WebRTC 栈改造聚焦‘薄中继+有状态 transceiver’架构以减少处理跳数
该优化已落地于 ChatGPT 语音交互与 Realtime API 生产环境

结构提纲

按章节快速跳转。

§问题本质
指出语音 AI 自然感的关键是对话节奏与人类言语同步，而非单纯音质或识别准确率。
·技术方案
重构 WebRTC 栈：引入轻量级媒体中继 + 有状态 transceiver，减少编解码与路由开销。
·落地场景
支撑 ChatGPT 语音模式与 Realtime API，实现毫秒级双向语音流响应。
›效果验证
生产环境实测显著降低端到端延迟，提升中断恢复与流稳定性。

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）

OpenAI 低延迟语音架构
- 核心目标
  - 对话节奏匹配人类语速
- 关键技术
  - 薄中继
  - 有状态 transceiver
- 应用场景
  - ChatGPT 语音
  - Realtime API

金句 / Highlights

值得收藏与分享的关键句。

🎙️ Voice AI only feels natural when conversation keeps pace with speech.
— 原文首句
⬇︎ 下载 PNG 𝕏 分享到 X
rebuilt our WebRTC stack with a thin relay and stateful transceiver
— 原文核心描述
⬇︎ 下载 PNG 𝕏 分享到 X
to keep real-time media fast for ChatGPT voice, the Realtime API, and more
— 原文应用范围说明
⬇︎ 下载 PNG 𝕏 分享到 X

#WebRTC#Voice AI#Realtime API#OpenAI#low-latency

打开原文

Here’s how we rebuilt our WebRTC stack with a thin relay and stateful transceiver to keep real-time media fast for ChatGPT voice, the Realtime API, and more.

https://t.co/JEvs2PmsmC" / X

OpenAI Developers on X: "🎙️ Voice AI only feels natural when conversation keeps pace with speech. Here’s how we rebuilt our WebRTC stack with a thin relay and stateful transceiver to keep real-time media fast for ChatGPT voice, the Realtime API, and more. https://t.co/JEvs2PmsmC" / X

Don’t miss what’s happening

OpenAI Developers ![Image 3](http://x.com/OpenAIDevs)

@OpenAIDevs

Image 4: 🎙️ Voice AI only feels natural when conversation keeps pace with speech. Here’s how we rebuilt our WebRTC stack with a thin relay and stateful transceiver to keep real-time media fast for ChatGPT voice, the Realtime API, and more.

How OpenAI delivers low-latency voice AI at scale

From openai.com

12:08 AM · May 5, 2026

41

94

880

452

Read 41 replies