GPT-Live、カスケード型音声パイプラインを廃止し全二重音声へ
OpenAIが「gpt-live」を発表。同時に聞き取りと発話を行うことでレイテンシを低減し、自然な割り込みを可能にする全二重音声モデルで、新しいChatGPT音声モードの基盤となる。
openai introduced gpt-live, a new voice model family built on a full-duplex architecture that listens and speaks simultaneously, ending the trade-offs of cascaded systems that chained stt, llm, and tts models. gpt-live can handle interruptions, back-channeling like 'mhmm', and delegate complex reasoning to gpt-5.5 in the background without halting the conversation flow. the rollout includes gpt-live-1 and gpt-live-1 mini for chatgpt voice, with api access coming later. this shift moves ai interaction toward fluid, human-like conversation suitable for complex, long-running agentic workflows.