(Bloomberg) -- OpenAI is letting businesses and developers include its real-time voice assistant in their own applications, paving the way for more users to have realistic-sounding spoken conversations with an artificial intelligence system for a wide range of tasks.
The feature, which is similar to the newest voice assistant available to paying users of its ChatGPT chatbot, is set to be made available on Tuesday to those building apps and services with OpenAI’s application programming interface, or API. Businesses will be able to develop software such as a voice-based customer service bot or a travel app that places phone calls on a user’s behalf, OpenAI said. Some companies have already been testing out the new voice capability, including health coaching app Healthify and language learning app Speak.The third-party voice integration is one of a number of updates OpenAI is announcing at a developer event in San Francisco on Tuesday. The conference offers OpenAI a chance to show how it plans to stay ahead of an increasingly crowded market for AI software at a time when it’s looking to close a large funding round. It also comes days after several leaders, including Chief Technology Officer Mira Murati, said they planned to depart the startup — the latest in a series of high-ranking employees to leave in recent months. While OpenAI unveiled a new AI model at its first-ever developer conference last November, the company said it won’t be doing so this year. Instead, OpenAI is focusing on showing off new API capabilities and meeting with some of the 3 million developers working with its software in different parts of the world. Tuesday’s event is intended to be the first of three, with others to be held in London and Singapore. At a press briefing ahead of the event, OpenAI employees demonstrated the new voice feature with a fictional travel app. Romain Huet, OpenAI’s head of developer experience, asked the fake app to call a made-up business, Ilan’s Strawberries, and order 400 chocolate-covered berries while keeping the budget under $1,500. Immediately, the app placed a call — a feature made possible, it was explained, by the app’s use of an API from cloud communications company Twilio Inc. — and a phone next to Huet rang. Huet picked up and, acting as if he was a worker at the strawberry business, took an order from the voice assistant. He said it would cost about $1,200 for the strawberries.
The voice used in the demo did not identify itself as an AI voice assistant — a lack of disclosure that might concern some users as well as privacy and digital rights advocates if applied to actual apps. OpenAI said after the demo it doesn’t mandate that developers inform users in any specific way when they’re interacting with AI. But the company pointed out that its usage policies state developers must generally tell users when they’re using AI, unless it’s made obvious by the type of exchange.
“There is no strict guideline of how you have to do it,” said Olivier Godement, the head of product for OpenAI’s API. “We just want to make sure it’s obviously clear to the user.”
OpenAI first teased a real-time voice feature in May for ChatGPT, showing how it could quickly talk back to users in response to written and visual prompts. But the next month, OpenAI delayed launching the option to work through potential safety issues. After the delay, OpenAI said the product would not be able to impersonate how other people speak. The company also said that it had added new filters to ensure the software can spot and refuse some requests to generate music or other forms of copyrighted audio.
In July, OpenAI offered the feature to a limited number of its ChatGPT Plus customers. Last month, the startup began rolling it out to all paying ChatGPT users.
©2024 Bloomberg L.P.
