AI GlossaryㅍTechnical words in the news
full-duplex
A communication method that allows simultaneous listening and speaking, enabling natural conversation with interruptions in voice AI
In plain words
Full-duplex is a communication method that lets both sides listen and speak at the same time. Think of a phone call: you can say "yeah" or "right" while the other person is still talking, and even if your words overlap, the conversation flows naturally. The opposite is a walkie-talkie, where one side has to say "over" before the other can start speaking. This one-direction-at-a-time style is called half-duplex.
This is exactly why older voice assistants felt frustrating. Most of them worked like walkie-talkies, running a separate step to detect whether the user had finished speaking before they could respond. If that detection was too quick, it cut the user off; if too slow, the conversation dragged. Full-duplex voice models remove this detection step entirely, processing listening and speaking through the same channel at the same time. As a result, the conversation can flow without breaking down even when speech overlaps or interrupts, just like a real human conversation.
How it shows up in the news
The article describes OpenAI's new voice system, GPT-Live, and calls its core feature full-duplex. A common misconception is that this is simply a technology for speeding up responses. In reality, it's a matter of structure rather than speed. By removing the step of first determining whether the other person has finished speaking before responding, it allows the system to react naturally without waiting for the other side to stop talking.
See also
Stories using this term
- OpenAI reveals design behind seamless voice AI GPT-LiveAI · 2026.08.04
- Google's ADK adds automated evaluation for live voice agentsAI · 2026.08.25
- ChatGPT Work now turns spoken thoughts into finished documentsAI · 2026.08.25
- ChatGPT rolls out bundle of updates including quiz generation and Google Drive integrationAI · 2026.08.15
- ChatGPT Work moves between PC apps and browsers to handle choresAI · 2026.08.29
- Base44 cuts token usage 20% after adopting GPT-5.6AI · 2026.08.18
