OpenAI re-launches voice feature

OpenAI has re-released its advanced voice feature (which is powered by its latest and most advanced AI model, GPT-4o and offers users hyper-realistic voice responses) to a small number of ChatGPT Plus subscribers.

After originally debuting the feature in May, OpenAI was quickly forced to take it down because actress Scarlett Johansson claimed they had used her voice to develop and train the pre-set voice of ‘Sky’, without her permission. 

Now, the feature has four pre-set voices (‘Sky’ isn’t available!) called Juniper, Breeze, Cove, and Ember, which have all been developed using paid voice actors. OpenAI claims that the voice feature can sense and respond to emotion and even allows users to interrupt, which offers them more natural, free-flowing, real-time conversations.

OpenAI has made it clear that it won’t impersonate public figure or individual voices, and will “block outputs that differ from one of these pre-set voices.”

The select few ChatGPT Plus users who have been chosen to test the voice feature will get a notification in the ChatGPT application followed by an email with instructions on how to use it. OpenAI plans to continuously add more users on a rolling basis, with plans to give access to all Plus users by the fall. 

They were meant to re-launch this feature in June, but delayed it by a month to improve “the model’s ability to detect and refuse certain content” and are rolling it out slowly, to a small set of people, to monitor its usage and help make the experience “safer and more enjoyable for everyone.” This is reassuring after hearing mounting reports from ex-employees, that they prioritize “shiny new product launches” over safety.