Washington, July 09 (QNA) - OpenAI has launched its new GPT-Live-1 voice model to run the voice chat feature in its chatbot ChatGPT to improve interaction with its smart assistant and make the conversation more natural and smooth, closer to natural dialogue between humans.

The company explained that the new model is its most advanced audio model to date, capable of listening ​and speaking ​simultaneously in real time. It includes improvements such as reduced interruptions, better understanding of natural pauses, and smoother responses.

It added that the new model can automatically use the best text models, such as GPT‑5.5, when it needs to infer or search the web, then return directly to the conversation to display the results without the user noticing an interruption in the dialogue.

The GPT-Live-1 model also allows the display of AI-generated visual information during conversations related to weather, stocks, or sports, such as match results or weekly weather forecasts, to provide a more interactive experience.

This technology differs from previous systems that operated in separate steps, in which the user's voice was converted into text, then the request was analyzed by a linguistic model, and finally an audio response was produced, resulting in delays and loss of crucial conversation context. The new model relies on a structure called "full duplex," which enables it to listen and speak simultaneously, processing inputs and producing responses synchronously and continuously.

This structure paves the way for new features, such as real-time translation during conversations, as users no longer need to pause their speech for ChatGPT to begin translating. Rather, the conversation is translated continuously, allowing the user to ask the assistant to remain silent until needed, providing subtle indications that it is listening, such as using short phrases to show it is ready to continue.

The company said that the new model includes built-in safeguards to limit harmful responses and provide age-appropriate responses for younger users. (QNA)