Builds the Voice Agent API's input.turn_detection block from the config bag.
AssemblyAI documents five fields there: vad_threshold (speech-detection sensitivity, 0–1, lower is more
sensitive), min_silence (silence before a confident end of turn), max_silence (silence that forces an end of
turn), interrupt_response and interruption_delay. Of MJ's normalized settings only Threshold has a field with
the same meaning, so it maps onto vad_threshold.
The rest are named in the returned warning:
SilenceDurationMs is one fixed trailing-silence window; AssemblyAI splits that job between min_silence and
max_silence around its own end-of-turn model, and either choice would change what the author asked for.
Eagerness and Mode have no field: AssemblyAI owns its turn model.
Coverage has no field: the session is audio only.
A raw turn_detection object in the bag is AssemblyAI's own vocabulary, so its fields win over the mapped ones.
Builds the Voice Agent API's
input.turn_detectionblock from the config bag.AssemblyAI documents five fields there:
vad_threshold(speech-detection sensitivity, 0–1, lower is more sensitive),min_silence(silence before a confident end of turn),max_silence(silence that forces an end of turn),interrupt_responseandinterruption_delay. Of MJ's normalized settings onlyThresholdhas a field with the same meaning, so it maps ontovad_threshold.The rest are named in the returned warning:
SilenceDurationMsis one fixed trailing-silence window; AssemblyAI splits that job betweenmin_silenceandmax_silencearound its own end-of-turn model, and either choice would change what the author asked for.EagernessandModehave no field: AssemblyAI owns its turn model.Coveragehas no field: the session is audio only.A raw
turn_detectionobject in the bag is AssemblyAI's own vocabulary, so its fields win over the mapped ones.