OpenAI's voice AI still relies on WebRTC, developer argues it drops audio
Developer kixelated, in a follow-up post after a viral critique, says OpenAI's voice products continue to use WebRTC for real-time audio, a protocol built for human conference calls that aggressively discards late-arriving packets rather than waiting for retransmission. He argues this causes voice AI systems to lose or misinterpret spoken input whenever there is packet loss, such as during a dropped wifi connection or cellular handoff. OpenAI engineers Sean DuBois and Justin Uberti reportedly reached out to him after his original post.
GoKawiil's interpretation of the reporting above, not reported fact.
The critique suggests that protocols designed for human-to-human calls may not suit AI voice interactions, where losing even a few words can produce incorrect responses rather than just a glitchy sound. If kixelated's technical argument holds, it could push AI companies to reconsider real-time transport choices for voice products as they scale. The direct engagement from OpenAI staff indicates the company is at least aware of the debate, though no public commitment to change was reported.
- WebRTC discards late audio packets by default, which the author says can corrupt voice AI transcription during network hiccups.
- OpenAI engineers Sean DuBois and Justin Uberti contacted the critic after his earlier viral post on the topic.
- The post argues retransmission features in WebRTC are disabled or ineffective for real-time audio jitter buffers.
Source: moq.dev, 2026-11-19
Published there as: “OpenAI still has a WebRTC Problem”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.