Is there a parameter to adjust STT confidence requirements E LiveKit Community #ask-ai

Is there a parameter to adjust STT confidence requ...

many-forest-60185

12/27/2025, 5:22 AM

Is there a parameter to adjust STT confidence requirements? E.g. The STT can only transcribe if the transcription is of a certain confidence

tall-airline-16756

12/27/2025, 5:22 AM

Thanks for reaching out, Alexander. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

tall-airline-16756

12/27/2025, 5:23 AM

Thanks — regarding your question about an STT confidence threshold: • Yes. Some STT plugins expose confidence-related parameters. For example, the AssemblyAI plugin provides

end_of_turn_confidence_threshold

(default 0.4) to control the minimum confidence required to consider a turn final. • Word-level confidence thresholds are usually controlled by the STT provider and may not always be exposed by every plugin. • If you need stricter/custom behavior, you can use pipeline nodes (for example

stt_node

) to inspect transcription results and apply your own filtering or acceptance logic based on the confidence values the STT provider returns. Would you like instructions for the AssemblyAI setting or an example of how to filter transcriptions in an

stt_node

? Sources: AssemblyAI plugin guide | LiveKit docs | Pipeline nodes and hooks | LiveKit docs

many-forest-60185

12/27/2025, 5:23 AM

What about for deepgram?

tall-airline-16756

12/27/2025, 5:23 AM

You're asking whether Deepgram has a configurable STT confidence threshold like AssemblyAI's

end_of_turn_confidence_threshold

. Deepgram does not expose a direct confidence-threshold parameter in the LiveKit plugin. End-of-turn/endpointing is handled by Deepgram's internal model logic and isn't tunable as a single confidence value in the LiveKit integration. Options you can use instead: • Adjust Deepgram's VAD-related parameters in the plugin (e.g.,

vad_min_silence_duration_ms

vad_speech_pad_ms

) to influence turn segmentation. • Post-process transcription results in an

stt_node

(or your application) to inspect any confidence scores Deepgram returns and apply your own acceptance/filtering logic. Would you like an example of how to filter Deepgram transcriptions by confidence in an

stt_node

, or the specific Deepgram plugin config fields to change? Sources: Baseten STT plugin guide | LiveKit docs | Deepgram STT plugin guide | LiveKit docs

Open in Slack

Previous Next