How GPT-4o powers GPT Live Voice Mode's full-duplex real-time translation

GPT Live uses full-duplex audio, letting the model listen and speak at once — like a phone call, not a walkie-talkie. Built on GPT-4o, which processes audio, text, and vision in a single neural network, it drops the older speech-to-text pipeline and preserves tone and emotional cues.