OpenAI updates ChatGPT with ability to hear and speak, see images, but it still cannot sing
- Users will be able to pick one of five voices in the ChatGPT app to produce audio responses to voice prompts
- The new features will be available to enterprise users and paid subscribers of OpenAI’s ChatGPT Plus service
Starting over the next two weeks, users will be able to choose a voice in the chatbot app, picking from five personas with names like “Juniper,” “Breeze” and “Ember”. ChatGPT will then produce audio of the text it generates in that voice – for example, reading an AI-generated bedtime story out loud.
The feature will be available to people who subscribe to OpenAI’s US$20-per-month ChatGPT Plus service and enterprise users.
OpenAI released its ChatGPT app in May, and already offers a voice-to-text capability that lets users talk to the bot. Adding an audio response feature could create a sense that people are having a more human conversation.
The company hopes the new feature will encourage on-the-go uses of its mobile app, putting it in closer competition with personal assistant offerings like Google’s Assistant, Apple’s Siri or Amazon.com’s Alexa.
Requests could include asking the program to talk about the history of Disneyland while driving to the theme park, or asking for a cocktail recipe while rummaging around in the kitchen. In a test of the tool, it ably narrated a story about a starfish and a swede.
However, while ChatGPT can come up with lyrics for songs, the app will decline to sing.
The voices of ChatGPT sound fairly human-like (though a close listen reveals a bit of a robotic monotone). OpenAI said it worked with voice actors to build the text-to-speech AI model that underlies the feature.
The company also said that in the coming weeks paid and enterprise users will be able to access a feature for GPT-4 – one of the AI models that powers ChatGPT – to submit a picture and a related question about it.
For example, it will be possible to upload a picture of pink sunglasses and ask the chatbot to suggest an outfit to go with it, or to submit a picture of a maths problem and request help solving it. The feature, which OpenAI announced earlier this year when it unveiled GPT-4, is available through the ChatGPT app and website.