OpenAI launched GPT-Live, a new family of voice models that can listen and speak at the same time, in real time.
Company said it will release two versions of GPT-Live for users worldwide. The models are called GPT-Live-1 and GPT-Live-1 mini and began to be distributed this Wednesday, according to OpenAI.
New models are full-duplex, which allows natural interruptions during the conversation. OpenAI claims they sound more natural and handle speech switching better, as well as enabling features like live translation.
ChatGPT will switch Advanced Voice Mode to GPT-Live-1 mini by default. Paid plan users will be able to access the larger model, GPT-Live-1, according to the company.
OpenAI says the new architecture fixes limitations of the previous voice mode. The old system combined a speech recognition model, a language model to generate responses, and a speech synthesis model to deliver the final audio.
Voice models now power OpenAI’s latest text models for specific tasks. The company stated that GPT-Live can send the query to models like GPT-5.5 when it needs search, reasoning or agent capabilities, without interrupting the conversation.
WHAT OPENAI WANTS WITH THE NEW VOICE
OpenAI bets that voice can become the main interface for more complex work on the computer. In a briefing, the company said the goal is to allow longer conversations and more “hands-free” use of the assistant.
ChatGPT Voice product lead reported that he has used the feature for long periods of time. Atty Eleti stated that he has had 30 to 40 minute conversations with the function during walks.
Eleti said the company sees voice as a path to longer-term agent assignments. “Over time, we think this will also unlock the ability to use voice as a kind of primary interface for computing and for managing increasingly complex and long-running agent jobs. The incredible use cases that we see people using Codex and ChatGPT to do, we think voice could be the interface of the future for all kinds of work,” Eleti said at a press briefing.
LIMITS, COMPETITION AND SAFEGUARDS
OpenAI says more than 150 million people use features like Voice and Dictation to speak with ChatGPT. Rivals like Apple and Amazon have also updated their assistants to improve conversations and context understanding, while startups are trying to make interactions more expressive.
Company states that it does not want to transform the resource into an AI “companion”. OpenAI says the models have safeguards for age-appropriate responses for teens and to offer resources if the conversation turns to topics like self-harm.
Demonstration showed that live translation still has limitations in some languages. In a test in Hindi, the assistant had a strong American accent and a tone considered unnatural; OpenAI said the mode is optimized for “most spoken languages”, without detailing which ones.
Source: www.noticiasaominuto.com.br
Source link
