As the adventures of OpenAI continue, you can now have a conversation out loud with ChatGPT. A function formerly reserved for paying subscribers and which is now accessible free of charge to everyone.
The last few days have been particularly eventful at OpenAI. While some employees considered resigning to join Microsoft, Sam Altman is ultimately expected to make a return, despite his previous ouster by the board. A real soap opera! But that doesn’t stop the company from continuing to update ChatGPT, far from it. After revealing the GPT, so that everyone can create personalized versions of the chatbot without having to code anything (see our article), she has just announced in a short video on X (formerly Twitter) that the function ChatGPT Voice, previously reserved for ChatGPT Plus subscribers, is now free for everyone in the mobile app, on both Android and iOS. A very simple to use option, which uses speech recognition and synthesis to allow you to speak to ChatGPT as you would with Google Assistant, Amazon Alexa or Apple Siri. Even better: the chatbot speaks French with a charming little American accent, while keeping a fairly natural voice – OpenAI offers several choices of voices, men or women, named Sky, Ember, Breeze, Juniper and Cove. Result: you really have the feeling of chatting with someone!
ChatGPT Voice: how does the voice assistant work?
Just like in writing, ChatGPT Voice can answer just about any question, generate content on the fly, clarify certain points, help with cooking or even tell stories. It does exactly everything the written chatbot does, but without you needing to use the app screen, with a helpful and friendly voice. Indeed, when voice mode is activated, you no longer need to use your hands: ChatGPT listens and waits for you to speak to it. You can even continue with another question, the chatbot being responsive. He keeps the conversation topics in mind throughout the discussion, which allows you to ask him to clarify his thoughts on a certain subject or to rephrase certain details of his answers. Note that the content of the exchange is fully transcribed in writing and saved in the application history, without the vocal exchange being recorded. We tested ChatGPT Voice by asking it about various topics, and the chatbot answered us quickly, without delay, with precision, in a fluid and natural language. To be honest, it’s even a little disturbing, we had the impression of chatting with a human being even more than in writing… On the other hand, unlike Google Assistant or Alexa, the AI is incapable to access data in your smartphone or your applications. You can’t ask it to perform actions for you, like launching a YouTube video for example.
To start a voice chat with the assistant, you need to go through the mobile app – the web version of ChatGPT doesn’t seem to have the feature integrated yet. Here are the steps to follow :
► If you haven’t already, download the ChatGPT app from the Play Store or App Store. Be careful to download the one developed by OpenAI, because the stores sometimes contain clones.
► In all cases, check that the application is updated on your smartphone.
► Launch the ChatGPT application.
► On the home page, tap the headphones logo at the bottom right of the usual prompt input field.
► A window opens to briefly introduce you to the function. Push the button keep on goingat the bottom of the screen.
► The application then asks you to choose the assistant voice that suits you best among Sky, Ember, Breeze, Juniper and Cove. A snippet will sound when you check a box – the voices are in English, but the chatbot speaks French. Select the one you prefer and press Confirm.
► A pop-up window then asks you for permission to access your device’s microphone. Press on When using the app.
► You can now start a discussion with the chatbot!
► To change the voice, simply go to your account settings, in the section Voice.
To understand what you’re saying, ChatGPT uses Whisper, its own open source speech recognition system, developed in-house from 680,000 hours of speech in multiple languages. This allows it to recognize voice requests in French, English and around forty other languages. The algorithm is even capable of juggling between languages within the same voice request. Impressive !