OpenAI is developing a bidirectional audio model, reportedly named BiDi, to facilitate natural and fluid AI conversations. This system processes speech continuously rather than using traditional turn-based methods.
The model adapts responses in real-time and handles user interruptions to mimic human interaction. This technology targets improvements in customer support applications and smart speaker devices.
Current prototypes are incomplete and reportedly malfunction after several minutes of continuous conversation. OpenAI may delay the release of the model, which was originally scheduled for the first quarter.