中断能力太弱，响应最快 3 秒，最慢则更长，无法直接用于对语音有高实时性要求的项目。

We also need to continue to improve the system. First, let's talk about interruptions. In environments where the test scenario is a bit more complex and there are many people talking at the same time, we need to interrupt at a very fast speed, then keep silent and continue with ASR continuous recognition until the appropriate time to disconnect, and then re-combine the content recognized earlier and send it to the large model again, so that the conversation with the large model does not seem abrupt.