Hands-On With ChatGPT's Advanced Voice Mode

OpenAI just rolled out Advanced Voice Mode on ChatGPT

Image by author

OpenAI started rolling out Advanced Voice Mode this week.

Not to be confused with the standard voice that ChatGPT has had for a long time. Advanced voice mode offers more natural, real-time conversations, can respond with emotion, and does a lot more.

I’ve been waiting for this feature since it was announced and, in this article, I’ll test out those mind-blowing demos that OpenAI showed us a few months ago to see if it lives up to expectations and I’ll also add some of my own tests to dive deeper into this new feature.

First things first—How to use it?

Advanced Voice Mode is available to ChatGPT Plus subscribers. This feature can be used on the ChatGPT mobile app (iOS and Android) and the macOS app.

To know whether you already have this feature enabled, open the ChatGPT app and see whether you have the button below.

When you tap on it, you’ll begin an advanced voice conversation. If you see the new blue orb below, you have this feature enabled.

OpenAI

In case you still don’t have it, you can try reinstalling the ChatGPT app. I quickly gained access to Advanced Voice Mode when I did that.

P.S. If you're in the EU, you'll need to use a VPN to unlock this feature

Is Advanced Voice Mode as good as in OpenAI’s demos?

One of the OpenAI demos that blew my mind was turning ChatGPT into a story narrator. Why? ChatGPT embodies characters with different personalities.

In the example below, I asked ChatGPT to embody a lion king, a small mouse, and an evil snake. The first sounded like a confident king, the second was a shy mouse and the third an evil snake with a very creepy laugh.

Try it yourself: I'm writing a story and I'm going to have you practice a couple voices with me for different characters. For the first one I'm thinking we're going to have a majestic lion- he's kind of an old King and I want you to say something like “who goes there” I really want you to embody this character

Advanced Voice Mode was as good as what OpenAI showed us a few months ago. In my other demos below, we’ll see that this feature doesn’t disappoint and we can verify that OpenAI didn’t make any video edits to its demos.

On the other hand, I was amazed by how natural the voice sounded. The voice I used for this first example (Cove) reminds me of audiobook narrators. Cove is described by OpenAI as composed and direct. In the next demo, I try Spruce which is a more calm and affirming voice.

ChatGPT can now respond with a wide range of emotions

For my second demo, I wanted to explore different emotions ChatGPT could do, so I asked it to be a commentator for a soccer match.

At first, it expressed excitement when commentating on a goal, then sadness when I asked it to commentate on a last-minute goal scored against its team (it even cried!). When I told it the goal was offside, it switched back to happiness and excitement. What amazed me the most was how it built up the excitement when I asked it to call the goal again.

Try it yourself: I'd like you to commentate a soccer match. Let’s imagine we’re in a Champions League final

As a side note, advanced voice can also recognize musical sounds like the strings of a guitar or when someone sings.

ChatGPT can be sarcastic and do accents

You can customize how you want ChatGPT talks. It can not only express a wide range of emotion but also be sarcastic. I also asked it to do different accents to spice up the sarcasm a bit.

Try it yourself: Hey, I'd like you to be super sarcastic. Everything you say from now on will be full of sarcasm. I’m thinking about moving to another country. I was thinking about the UK. Do you think it’s a good choice? Answer me with a British accent.

I have to say I’m not an expert in accents and I’m not from any of the countries I mentioned in the video, so, please, let me know whether you think ChatGPT does a good British/Aussie/American accent.

ChatGPT can still make mistakes in Voice Mode

While Voice Mode adds new capabilities to ChatGPT, its flaws remain, so be mindful of its responses. I asked ChatGPT to count from 1 to 10, both slowly and quickly, but when I challenged it to count odd numbers in Spanish and even numbers in English, it failed multiple times.

Now it’s your turn to try Advanced Voice Mode! Just remember that there’s a daily limit (you’ll be notified when you have 15 minutes left of advanced voice for the day)