Turn-detection simulator — VAD ➜ Smart Turn v3.2

Replays your audio through the exact Silero VAD state machine and Smart Turn buffering used in a live voice pipeline. When VAD detects the user stopped speaking (stop_secs of silence), the buffered segment is sent to Smart Turn — the same trigger used live.

Each fired end-of-turn appears in the table below.

Audio (upload or record)

Model

VAD (Silero)

confidence

0 1

start_secs (speech-start confirm)

0 1

stop_secs (silence before VAD stop → triggers Smart Turn)

0 1

min_volume

0 1

Smart Turn

pre_speech_ms (lead-in)

0 2000

max_duration_secs (model window cap)

1 16

stop_secs fallback (force end-of-turn on long silence)

0.5 10

decision threshold

0 1

Simulation

chunk_ms (transport frame size)

10 100

append trailing silence (ms)

0 2000

Also run at end of recording (force a prediction on the full audio even if VAD never reports a stop)

Last end-of-turn decision

End-of-turn events

Details

Turn-detection simulator — VAD ➜ Smart Turn v3.2

🔊 Audio passed to Smart Turn (per detected turn)