ChatGPT Advanced Voice has arrived for most Plus users

After a longer than expected wait, the Advanced Voice mode for ChatGPT has finally rolled out to the majority of ChatGPT Plus subscribers. The voice feature in ChatGPT already allows you to interact with ChatGPT with just your voice, and have the responses spoken back to you. The updated voice feature now allows you to talk to the ChatGPT app in a much more natural way, with almost instant responses and even more natural voices.

The feature was first teased way back in May, and users have eagerly awaited its release, with a lucky few getting early access over the past weeks. With the feature finally being rolled out to all ChatGPT Plus users, I took Advanced Voice for a spin to see if it lives up to its promise.

The new voices give you far more choice

Sky may be gone, but there are some good alternatives

It’s fair to say that despite all of its success, OpenAI has made a few missteps along the way. One of the most dramatic was the creation of a voice for ChatGPT that sounded more than a little like the actress Scarlett Johansson, who played the voice of an AI chatbot in the movie Her. Apparently, OpenAI did approach Johansson about using her voice, but when she refused, they used one that sounded very like her anyway. Thanks to her lawyers, the Sky voice is no longer available, which sucks because it was my favorite by far.

Thankfully, we have four new voices now, joining the pre-existing Ember, Juniper, Cove, and Breeze. Arbor is a male British accent that sounds a little like Karl Urban’s slightly odd cockney accent in Amazon Prime’s excellent show The Boys. Arbor doesn’t quite have the same vocabulary, however. Spruce is a “calm and affirming” voice that has slight vibes of Morgan Freeman. Sol is a female accent with a hint of valley girl, and if you ask her to speak like a valley girl, she really can.

My favorite voice, however, is Vale. This is a female British accent, and it’s like having your own version of Mary Poppins to chat to. The voices are all excellent and give you far more choice than we’ve had previously.

Response times and tone of voice are excellent

It really does feel like a natural conversation

This is the biggest change with Advanced Voice; the long delays waiting for a reply have completely gone. No longer do you have to watch the on-screen animation as ChatGPT thinks about a reply; the response is instant, just as if you were speaking to another person.

It really does make a huge difference. Talking to ChatGPT had never truly felt like a genuine conversation before, because of the long unnatural pauses, but now the back and forth is so instantaneous that it’s easy to forget you’re talking to software.

The ability to interrupt also helps; you can speak while ChatGPT is talking, and the speech will stop as the app listens to what you’re saying. It’s especially useful if ChatGPT is giving a long response that’s off-topic, and you want to get things back on track. It, too, feels very natural, if a little rude, as you’re effectively talking over the top of someone else.

The ability to speak in different types of voice is impressive

ChatGPT can do a better troll voice than I can

Another really impressive ability of Advanced Voice is the way that all the voice options are able to change the tone or style of their voice to meet your specific requests or just to fit the context of what’s being said. After I asked ChatGPT to tell me an original joke, the response was awful, and I told ChatGPT so. ChatGPT replied saying “Yeah, that one really landed in Dad joke territory!” but the way the sentence was phrased, you could almost imagine someone speaking and pulling a jokey, embarrassed face.

Even more impressive was when I asked ChatGPT to tell me the story of the Three Billy Goats Gruff and to use different voices for each of the characters. ChatGPT used a gentle, high-pitched voice for the smallest goat, and a deep, booming voice for the troll. It was right on the money, and honestly felt like I was listening to a real person tell the story.

ChatGPT hasn’t got any smarter

There are still two Rs in strawberry, apparently

While the voice features of ChatGPT have taken a big leap forward, ChatGPT’s smarts haven’t. I had quite a lengthy conversation with ChatGPT voice to try and get it to give me the correct answer to a simple question that famously stumps the majority of AI chatbots: how many times does the letter R appear in the word strawberry? Almost all AI chatbots answer two to this question.

Asking Advanced Voice, I got the same incorrect response. I tried getting ChatGPT to spell out the word one letter at a time and then counting the Rs, and it still said two. I asked it to spell out the word one letter at a time again and add one to the total every time the letter was an R, and it almost worked, but it missed out some of the letters of strawberry as it was counting. Finally, after getting it to try again, it managed to get an answer of three.

OpenAI’s new o1 model, which is capable of far more complex reasoning, is able to answer this question, although it still takes several seconds to do so. However, you can’t use o1 with voice; it will always revert to GPT-4 or GPT-4o which are less capable of reasoning.

Some Plus users still can’t get Advanced Voice

The EU, UK, and others will have to keep waiting

Not everyone with a ChatGPT Plus subscription has been given access to the new Advanced Voice feature. On the FAQ page for the new feature, it states that Advanced Voice is not yet available in the EU, the UK, Switzerland, Iceland, Norway, and Liechtenstein. Iceland, Liechtenstein, Norway, and Switzerland are part of the European Free Trade association that operates in parallel with the EU, and the UK was part of the EU until 2020, so it is highly likely that this relates to EU privacy laws. The word “yet” does seem to indicate that the feature will come to these countries at some point in the future.

Some users in these countries have reported being able to access Advanced Voice through the use of a VPN, although it’s unclear whether this violates any terms of service. It is understandably frustrating for users in these countries, since it was not mentioned at any point prior to the release of Advanced Voice that these countries would not have access, and many users from these regions will have been paying for their ChatGPT Plus subscriptions for exactly this reason.

ChatGPT is still lacking some of its best features

This is a big leap for voice but we’re still waiting for video

OpenAI

There’s no denying that Advanced Voice is a big leap forward for voice interaction in ChatGPT. Using the new feature genuinely feels like having a conversation with another person, thanks to the instant responses and the ability to modulate the tone of voice to suit the context.

When researching how people use ChatGPT, I found that many people use it as a source of emotional support, as you can talk to ChatGPT without judgment. The new Advanced Voice feels so much more like a real person, and this means it’s even more likely to be useful for people who are just looking for someone to talk to.

It’s taken a long time for us to reach this point, however. OpenAI gave a demo of the Advanced Voice feature all the way back in May, with the promise that the feature would be rolled out “in the coming weeks.” It’s now more than four months later and Advanced Voice has only just arrived.

We also saw some even more impressive demos back in May, including the ability to use a live feed from your camera to ask ChatGPT about what you can see all around you. It seems very likely that we won’t be seeing that feature any time soon.