I’ve been using ChatGPT’s advanced voice model since it was activated on my Plus account, several months after its official introduction by Open AI. Until now, I had to use the VPN thing, with the latency in responses that the tunneled connection adds. But you don’t have to do that anymore: ChatGPT’s advanced voice assistant is officially available in Europe.
The parallel between the film Her and the advanced voice assistant of ChatGPT is not a coincidence, Open AI itself was inspired (more than necessary) by the famous film by Spike Jonze. And since I tried it I can only agree, it is very easy to forget that you are talking to a machine. The inflections of the voice, the laughter, the breathing… The magic of the details is what ultimately cements the human experience. And now that I no longer need to use a VPN, I can use ChatGPT as my personal assistant.
ChatGPT’s advanced voice model now works in Europe
As I said, I’ve been using the new voice model and chatting with it since it was activated in my ChatGPT Plus at the end of September. To use it I had to first activate my VPN so that the IP address pointed to the United States. Then the assistant would move from the old environment, which was already very good, to the new one, which was excellent. This is no longer necessary, as Open AI itself announced.
What makes advanced AI great is that its conversational mode has almost no latency (it responds like a person would after listening to you), Open AI models perfectly adapt language on demand, and , above all, the voice builds a personality it’s almost human in the details. At first this returns a certain mechanical tone, but this can be changed by asking. After my tests, the Andalusian accent is the one that seems the most natural to me.
The advanced audio model has nine voices. My favorite is Vale with a change of Andalusian accent: it’s the best natural recipe I’ve found
A curiosity is that the voice assistant depends on the selected language model. For example, if I choose o1-preview, the argument is much more complex, but the reflection takes several seconds; which ends up hindering the conversation. With o1-mini it’s a bit faster, but it’s too similar to the old voice. GPT-4o is perfect.
I can ask it to explain all my questions, start a random conversation, ChatGPT lets me model voice behavior on demand, helps me with documentation tasks when I can’t take my eyes off the keyboard and, no less important, it amuses me. Because he has a lot of ingenuity.
By imitating it, I was able to chat with the AI so that it would tell me jokes, to solve stories, it’s great for practicing English (sometimes I ask it for a “talk” session so I don’t (not get rusty) and I tested it on two simultaneous phones to see how far it goes in the conversation. Once a whole language was invented by talking to herself and adding new words. And it amazes me how he creates a story by throwing challenges from mobile to mobile.
ChatGPT’s advanced voice is very good, but it’s far from perfect
Open AI has focused on naturalness, I have already highlighted the fact that the details of the voice are what ultimately give humanity. Language models do their part of analysis and response themselves the voice says what the AI writes. And being able to interact with ChatGPT without looking or touching your phone is wonderful. Now, good.
Although it works perfectly, the new voice model is far from what Open AI promised: it cannot analyze the environment with the camera, it does not perform mathematical analysis when focusing on a problem with the cell phone and, in general, it does not allow interacting with the environment as if it had virtual eyes. Everything about the camera still doesn’t work, which detracts from the promised enormous potential.
Open AI has not yet introduced all real-world interactions into the voice model, taking advantage of the mobile camera
Another point is that, although I can use ChatGPT as a personal assistant, its ability to manage my mobile phone, or my connected objects, is zero. I would love to be able to tell him to turn off the WiFi, disconnect my house alarm, or check the cameras to tell me if he sees anything strange, but no. And I doubt Open AI will introduce it in the future, even if it is technically viable: the app could interact with mobile hardware and software using Android APIs.
Cover image | DALL-E 3 in ChatGPT Plus modified
In Xataka Android | All mobile phones updated to Android 15 and when will they start updating
In Xataka Android | How to share your Android data connection with other devices
