The LexiconRuntime component has several settings that you can tweak to improve the experience. If you change any of these settings at runtime call the LexiconRuntime Restart() method for the changes to take effect.

Speech to Text

Speech Confidence Threshold	The confidence value required for speech to text results to be processed (value between 0 and 1)
Use Custom Language Model	Whether to use the custom language model for speech to text (only available if you trained a custom model)
Custom Language Model Weight	How much to favor the custom language model over the base language model (value between 0 and 1)
Keyword Confidence Threshold	The confidence value required for a keyword match (value between 0 and 1)

Assistant

Intent Confidence Threshold	The confidence value required for an intent to be processed (value between 0 and 1)
Match Multiple Intents	Whether to allow multiple intent matches. Leave unchecked to only process the best intent match

Calibration

Silence Threshold	Microphone input levels below this value will be considered silence (value between 0 and 1)
Timestamp Offset	Adjustment for timestamp values (in seconds). Use to improve the alignment of uttered words with focus data

Future versions of Lexicon will provide calibration tools for silence threshold and timestamp offset. For now, manually adjusting these values for a particular device and user is the best approach.