breezy-knife-25008
11/13/2025, 4:04 PMsession.say() instead of session.generate_reply(tool_choice="none") and that did clean up the initial greetings, but probably because they're hard coded now.
For context this is with the restaurant starter recipe that was converted to use inference.
This issue happens locally without SIP, and when calling livekit cloud sip / telnyx.
I am just wondering where to optimize or fix here.
Could it be related to using inference, or maybe VAD, or is it the actual STT and TTS models?
I was also curious if it could be because I moved my prompts over to separate yaml files. Claude code was saying that's not the culprit.
Anyone else having similar issues?
If so any luck solving?