Hi! I am using realtime API with websockets and “semantic_vad” turn detection. The turns are detected accurately, but the model responds to the previous to last question. Here is an example:
Me: Hi! What date is it today?
Model: Hello, how can I assist you?
Me: Did you hear my question?
Model: Today is March tenth.
Me: And what time is it?
Model: I apologize! Today is March tenth.
Is it a known problem, or something on my side?