OpenAI opens its speech AI engine to developers

Driving the news: The new speech-to-speech feature is one of several announcements that OpenAI is making at Tuesday's DevDay event in San Francisco.

• Early testers of the feature include nutrition and fitness app Healthify and Speak, a language learning app.
• Other new features being made available to developers include the ability to fine tune models based on pictures.

Zoom in: In a demo for reporters, OpenAI executives showed an example of the new audio capabilities combined with Twilio's API to allow an AI assistant to call a fictional candy shop and place an order for 400 chocolate covered strawberries.

• Developers will only be able to use the voices provided by OpenAI — the same ones that are options within ChatGPT.
• While the voice won't be watermarked in any way and developers won't have to make the AI system identify itself, OpenAI says it's against the company's terms of service to use its systems to spam or mislead people.

The big picture: The announcements come amid a wave of news around the ChatGPT maker, including its ongoing massive fundraising effort and last week's departure of CTO Mira Murati and two other top executives.