OpenAI said on Thursday it has updated ChatGPT's voice feature based on the GPT-Live-1.1 (GPT-Live-1.1) voice model.

The update improves intonation and vocal expression in multiple languages, including Korean, to help users hold more natural conversations.

It also added two new voices for Korean users: the male voice "Hanul" and the female voice "Haetsal". Users can choose a preferred voice to ask questions or use it for foreign-language conversation practice.

It also improved how conversations continue during information searches. Instead of repeatedly saying "checking" as before, it uses a variety of phrases suited to the situation so users can keep talking naturally while waiting for an answer.

ChatGPT Voice is based on a full duplex structure that lets users speak while listening to the other party's voice at the same time. It can respond when users ask an additional question during ChatGPT's reply or pause briefly to gather their thoughts.

OpenAI introduced this voice conversation method when it launched GPT-Live in July, and further improved multilingual voice quality and conversational phrasing in this update.

The update will be rolled out sequentially to ChatGPT users worldwide, including South Korea. It will be available to free users, who can select the voice icon in ChatGPT.