OpenAI announced on Tuesday that it is rolling out Advanced Voice Mode (AVM) to a select number of its paid ChatGPT customers. The voice feature, which makes conversations on ChatGPT more natural, will first be rolled out to ChatGPT's Plus and Teams tier customers. Enterprise and Edu customers will have access to it starting next week.
As part of the rollout, AVM will have a new design: The feature is now represented by a blue animated sphere rather than the black animated dot that OpenAI unveiled at a showcase of the technology in May.
Once AVM is available, a pop-up will appear next to the audio icon in the ChatGPT app.
Advanced Voice will be rolling out to all Plus and Team users of the ChatGPT app later this week.
While you all patiently waited, we added custom instructions, memory, five new voices, and improved accents.
You can also say “Sorry I'm late” in over 50 languages. pic.twitter.com/APOqqhXtDg
— OpenAI (@OpenAI) September 24, 2024
ChatGPT also added five new voices for users to try out: Arbor, Maple, Sol, Spruce, and Vale. This brings ChatGPT's total number of voices to nine (almost the same number as Google's Gemini Live), with Breeze, Juniper, Cove, and Ember. You may notice that these names are all inspired by nature, which could be because AVM's overall aim is to make ChatGPT more natural to use.
One voice missing from the lineup is Skye, a voice that OpenAI unveiled in its spring update that led to legal threats from Scarlett Johansson. The actress, who played an AI system in the feature film Her, claimed that Skye sounded a little too similar to her own. OpenAI quickly removed Skye's voice, saying it had no intention of sounding like Johansson, even though several staff members had referenced the film in tweets at the time.
Another feature not included in this rollout is ChatGPT video and screen sharing, which OpenAI debuted in its Spring update four months ago. The feature would enable GPT-4o to process visual and auditory information simultaneously. In a demo, OpenAI staff showed how to ask ChatGPT real-time questions about math questions written on a piece of paper in front of them or code on a computer screen. At this time, OpenAI has not provided a timeline for when it will release these multimodal features.
Screenshot image credit: OpenAI
That said, OpenAI says it has made some improvements since releasing AVM in limited alpha testing. ChatGPT's voice feature is said to be better at understanding accents, and the company says conversations are smoother and faster. Glitches were found to be common during testing of AVM, but the company claims they have been fixed.
OpenAI is extending some of ChatGPT's customization features to AVM: Custom Instructions allow users to customize how ChatGPT responds, and Memory allows ChatGPT to remember conversations for later reference.
An OpenAI spokesperson said AVM is not yet available in some regions, including the EU, the UK, Switzerland, Iceland, Norway and Liechtenstein.