The leaked GPT-4o voice included unique sound effects such as thunder and footsteps, which were tied to the story it was telling. This feature showcases the model's ability to generate sound effects in addition to its advanced voice capabilities.
RozziTheCreator accidentally gained early access to the GPT-4o feature when it suddenly appeared as an option while they were using ChatGPT. OpenAI later confirmed that a small number of users were given access to the model by mistake during testing. The feature has since been corrected and is not yet available to the wider public.
During RozziTheCreator's GPT-4o experience, they accidentally gained access to the advanced voice feature, which was not yet supposed to be available to users. They were able to use the voice to tell a horror story with sound effects, but the experience was very buggy and the AI started repeating and replying to things that weren't said. Access to the feature was quickly revoked.