According to CoinMeta, Tavus has launched a real-time dialogue understanding model called Sparrow-2, which is specifically designed to determine when in a voice interaction it is appropriate to listen, wait, speak, or continue speaking. This model updates the dialogue state every 10 milliseconds and is capable of handling common issues in traditional voice interactions, such as continuing to wait after a user pauses and recognizing encouraging tones. Tests have shown that the failure rate of this model's dialogue timing is only 2.1%, and in cases where the user pauses for more than 1 second, the system will continue to wait 97% of the time. The median response time of this model is 680 milliseconds, which significantly enhances the patience and judgment capabilities in dialogue interactions.