For services that don't emit UserStarted/StoppedSpeakingFrame (Nova
Sonic, Gemini Live, Ultravox), the absence of those frames means
downstream consumers — including the Pipecat Prebuilt UI — can't group
user transcripts into discrete turns. The Tier 1 comment block already
called this out, but the fix required users to know to add the
SileroVADAnalyzer import + LLMUserAggregatorParams kwarg themselves.
Make it a copy-paste: include the relevant imports and `user_params=`
argument as commented-out code, with a comment explaining that they're
not strictly necessary for context aggregation but enable RTVI / turn-
dependent processors when needed. Mirror the wording used in the
LLMService startup log.
Also fix line wrapping in the llm_service.py startup log for the no-
turn-frames case (manual edit to that message left the last line over-
length).
86f9ad0c07