Skip to content
Trending storyWatching

The WebRTC dispute over OpenAI's voice AI

1 report1 sourceupdated yesterday

What happened

AI digest

On November 19, Hacker News' front page carried an article arguing that OpenAI's voice AI products have a WebRTC problem. The author says WebRTC was designed for real-time human meetings and by default drops late audio packets without retransmitting them, which makes voice AI lose context and waste tokens. He proposes MoQ instead, saying one second of Opus audio at 64Kbps is only 8KB, which allows retransmission windows of up to 5 seconds so the model has full context before it needs it. He also argues WebRTC renders audio by arrival time rather than timestamp, which hurts TTS playback timing. The piece is the author's technical analysis and proposed alternative.

Written by AI from the coverage · updated yesterday

Coverage

Follow the reports to see the story from different sides.

Nov 19
  1. Hacker News front page
    OpenAI still has a WebRTC Problem

    针对 OpenAI 语音 AI 产品使用 WebRTC 的问题,作者指出 WebRTC 为人类实时会议设计,默认丢弃迟到音频包且不重传,导致语音 AI 丢失上下文、浪费 token。他提出用 MoQ 替代:音频 1s Opus @ 64Kbps 仅 8KB,可重传最长 5s,让模型在需要前收齐上下文。WebRTC 还按到达时间而非时间戳渲染音频,不利于 TTS 按时播放。

Heat over time

Heat now 5·Comparable peak 10(Nov 20 00:00)·Comparable change over 24 hours –

02.557.510Nov 2000:00Nov 2008:00Nov 2015:00Nov 2023:00

The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.