OpenAI has begun rolling out access to the "advanced" voice mode in ChatGPT, which it had been testing with a "very limited" number of users. This feature will be available to subscribers of paid plans.

Plus and Team users will be able to try the Advanced Voice mode during the week ending on September 29, 2024. Subscribers of the Enterprise and Edu plans will gain access the following week, the company announced. Those who get access will see a welcome notification. In this notification, OpenAI states that the voice assistant can respond to humor and interruptions, and it can base its answers on information from the "Memory" feature (which allows the bot to remember or forget things upon request). The developer also warns that errors are possible in the responses, so it’s best to double-check important details.

Advanced Voice is "currently" unavailable in the UK, EU countries, Switzerland, Iceland, Norway, and Liechtenstein. Users in these regions are encouraged to try using a U.S. IP address.

The welcome notification

Five new voices have been added to the mode. Additionally, improvements have been made to accents in certain languages, and the assistant's speech speed has been increased. Audio dialogue recordings are saved, but users can delete them at any time.

The voice assistant demonstrates how to apologize to a grandmother for being late — in English and Chinese.

OpenAI showcased the voice mode's capabilities during the GPT-4 presentation in May 2024. During the demonstration, speakers could interrupt the chatbot mid-response to ask for clarification or request it to tell a story in a special voice or even sing it.

The alpha version of the mode was supposed to be released at the end of June 2024, but OpenAI postponed the launch for further refinement. In July, it began testing Advanced Voice with a "very limited" number of users.