OpenAI launches GPT-Live with full-duplex architecture for real-time voice

OpenAI has introduced GPT-Live, a new generation of voice models designed for natural, bidirectional human-AI interaction. Unlike previous cascaded or turn-based systems, GPT-Live utilizes a full-duplex architecture that allows the model to listen and speak simultaneously. This design enables the AI to handle live interruptions, process background noise more effectively, and provide verbal cues like "mhmm" to signal active listening. Source

Læs hele artiklen hos kilden.