AI Technology That Can Listen and Speak at the Same Time Like a Human Is Being Developed

Russian company “Sber” is developing a new artificial intelligence technology known as a “full-duplex” system. The technology is designed to listen and speak at the same time, much like a person during a phone conversation.

Sergey Markov, Sberbank’s director for AI technology development, said this in comments to the “RBC” news agency.

At present, voice AI assistants usually operate in turns: a person speaks, the AI listens and processes the information, and then provides a response.

The main difference of the new technology is that it uses a single model instead of several neural networks. In existing systems, one neural network converts human speech into text, another generates the response, and a third turns that response back into speech. “Full-duplex” technology, by contrast, is designed to listen and speak simultaneously.

Similar technologies have already begun to appear around the world. One of the first was “Moshi,” introduced in 2024 by French startup “Kyutai.” Later, “Google” launched “Gemini Live,” while “OpenAI” introduced its “GPT-Live” solution.