ChatGPT's new voice mode mimics accents, sings, and corrects pronunciation

ChatGPT can now talk like an airline pilot over an intercom, laugh at its own jokes in Turkish, and break into a mournful blues version of Happy Birthday.

OpenAI has officially kicked off the rollout of its much-anticipated Advanced Voice Mode, powered by GPT-4o. But for now, access is locked behind an exclusive Alpha test sent via email to a select group of ChatGPT Plus subscribers.

Accents, emotional range, and real-time corrections

The early clips popping up on social media show that the conversation latency has practically vanished.

In one demo on X, user @nickfloats asked ChatGPT to tell a story like an airline pilot addressing passengers. The bot jumped in within a second, altering its vocal output to mimic the crackle of a cabin intercom. Layering actual engine roar over the voice was still a step too far, but it took user interruptions seamlessly.

Other clips highlight its flexibility across dozens of languages. In one video, ChatGPT corrected a user's French pronunciation by offering precise tips on vocal inflection. In another, it narrated an emotional story in Turkish, laughing and crying on cue, though native speakers noted the accent did not sound fully authentic.

The model can also pivot across regional US dialects, from Boston and New York to Wisconsin and a classic valley girl accent. When asked to sing, it produced a blues rendition of Happy Birthday and even attempted to mimic a cat and a frog singing the tune.

Several male and female voices are included in the preview, though the controversial "Sky" voice that OpenAI pulled in May remains missing. Video capabilities and screen sharing shown during the spring demo are also absent for now and will arrive later.

OpenAI spokesperson Taya Christianson confirmed that the company plans to bring Advanced Voice Mode to all Plus members, who pay $20 a month, sometime this fall.

Why it matters

For everyday users, this marks the shift from texting a machine to actually conversing with one. Language learners gain an interactive tutor that corrects nuances on the fly, and brainstormers get a sparring partner that senses hesitation and enthusiasm. That said, we don't think it is worth upgrading to Plus just yet, since the Alpha rollout is tiny and lacks the screen-sharing tricks shown in the original demo.

For the tech landscape, the bar for voice assistants has dramatically changed. Google and Apple now have to deal with an AI that doesn't just read answers off the web, but responds emotionally and tolerates mid-sentence interruptions.

We think OpenAI's cautious rollout is the smart play here. Real-time audio comes with huge compute costs and serious safety headaches around voice imitation. Stripping out the controversial voice and testing in a closed Alpha lets them iron out the weird edge cases before opening the floodgates.