Your issue occurs because only interim (partial) Speechmatics STT results are being processed, not the finalized transcriptions. To ensure the LLM receives finalized text, follow these steps: • Process Only Final Transcript Events: Verify that your implementation listens for the

FINAL_TRANSCRIPT

event, not just

INTERIM_TRANSCRIPT

. Only send final results to the LLM. • Event Handlers: Check that your

on_final_transcript

handler is registered and properly emits or forwards finalized text for downstream processing. • Turn Commitment: Ensure your solution calls

commit_user_turn()

and waits for final transcripts before fallback to interim. • STT Configuration: Confirm that your Speechmatics instance and turn detection config (VAD/manual) are set up to produce and handle final events. • Timeouts: Adjust transcript finalization timeouts to allow final events to arrive before falling back to partials. This setup ensures only finalized transcriptions are sent to the LLM, preventing the use of interim results in LLM responses. Let me know if you need help checking your event handling or configuration details. Sources: Speech-to-text (STT) integrations | LiveKit Docs | livekit-plugins/livekit-plugins-gladia/livekit/plugins/gladia/stt.py | LiveKit turn detector plugin | LiveKit Docs