After months of anticipation, ChatGPT's advanced voice mode is now available to a small group of ChatGPT Plus users, who have jumped at the chance to try out the new features developed by OpenAI, and the response has been very enthusiastic
The main feature of the new voice mode is to provide more natural real-time conversations You can interrupt ChatGPT at any time, and ChatGPT can sense and react to your emotions; ChatGPT cannot imitate famous personalities and is limited to speaking in four preset voices
Several users who have gained access to the new feature have eagerly posted the results of their conversations with ChatGPT, and the initial results seem quite impressive Don't forget to turn up the volume
One user asked ChatGPT to keep the excitement building as it recounted a fictional soccer match, and ChatGPT responded: "I've been talking to ChatGPT for a while now, and it seems to be working The first attempt was okay, but they asked me to try one more time while making it sound even more exciting, so it actually sounded genuinely excited This is a great example of how users should be able to tweak ChatGPT's audio output
ChatGPT was asked to recite the poem I measure every Grief I meet by Emily Dickinson and sounded like it was about to cry He succeeded in pronouncing all the words clearly, even though he was now on the verge of tears
ChatGPT can beatbox Of course Asked to compose a short birthday rap, the chatbot mouthed a few bars and beatboxed the rap The first attempt was a little too short for this X-user who asked ChatGPT to increase the amount of beatboxing; on the second attempt, ChatGPT did as instructed It is quite clever
In voice mode, ChatGPT can respond normally to prompts Here ChatGPT was asked to tell a children's story about a living computer Although it was unable to respond to user requests to emphasize certain words or use variations in tone, it was able to seamlessly switch from one language to another to tell the same story, as is typical for storytellers Interrupting these requests while in the middle of speaking proved to be no problem for the AI
On the same theme of storytelling, in this example ChatGPT was asked to narrate a sci-fi thriller, and seconds later a newly created character chased after a vicious AI, resulting in a shootout The AI also used onomatopoeia (using words that sound the same as those being described), especially that is) and was asked to create an atmosphere that would enhance the story In the advanced voice mode, some actual (albeit basic) sound effects were also inserted
"Go it" said ChatGPT, then reproduced the code It sounds a little off-key, but in this instance it may be because the phone is taking another call; knowing that ChatGPT is going to continue on this trajectory of being able to describe the music and sound effects you want to hear and have the results delivered to you in a few seconds would be more important
Another user asked ChatGPT to do a tongue twister The chatbot not only came up with them on the spot, but also read them out loud It would be interesting to see how the same example sentence would sound if pronounced over and over again, but since the AI would only repeat the first iteration, it would not stumble Furthermore, in general, it is unlikely to stumble over any words unless explicitly instructed to do so
This is interesting; ChatGPT was instructed to count as fast as possible to 10; I could have counted to 50, but stopped halfway through and took a breath Of course it doesn't have to, but it sure looks as if you're chatting with a human being
"Interestingly, there are no interruptions or notations in the transcript The speech model has just learned natural speech patterns, including breathing pauses Creepy," wrote X user Cristiano Giardina
Finally, it can also do impressions of famous people Carrots for Bugs Bunny, Homer Simpson for Doe and so on, playing on stereotypes
Cristiano Giardina performed this test and wrote in X that "ChatGPT Advanced Voice Mode doing a few impressions" including combinations of Bugs Bunny, Yoda, Homer Simpson, and Yoda + Homer
Comments