I’m experimenting with the Agents SDK (TS) with Realtime, and my use-case requires a tool call in place of voice response. I’m finding I always receive a voice response along with the tool call, regardless of prompting. I’d rather not resort to snubbing the voice output before it goes across the wire (websockets in my case) as that would be a bit hack-y, so I’m curious about any existing options or conventions around this.