Detects prolonged customer silence during a Virtual Agent (VA) conversation and reports it as a "VA Silence Detected" event, for consumption by the VA Sync node.
Parameters
name |
description | default |
|---|---|---|
| Silence Duration Threshold(ms) | How long the customer must stay silent, in milliseconds, before a "VA Silence Detected" event is raised. | 5000 |
Inputs
Audio
Any audio reaching this node while the customer is considered silent is counted towards the silence threshold.
Events
| name | description | possible nodes this event may be received from |
|---|---|---|
| Speech Started | Resets the accumulated silence length and marks the customer as no longer silent. | VAD Elastic, Silero Vad, Vad Service |
| Speech Ended | Marks the customer as silent, starting audio-based silence accumulation. | VAD Elastic, Silero Vad, Vad Service |
| SR Unrecognizable Speech | Reported immediately as a "VA Silence Detected" event, using Data.DetectedContent as the silence type (e.g. noise, dtmf). Also resets the accumulated silence length. |
SR Http, SR Websocket |
Silence detection based on the audio-driven threshold (Speech Started/Ended + accumulated audio) and the SR Unrecognizable Speech shortcut are both gated behind the VaSilenceCheckEnabled runtime/session parameter: while it is false, incoming audio and SR Unrecognizable Speech events are ignored and no silence event is produced. Turning it from true to false also clears any silence length accumulated so far. This parameter is typically toggled by the VA TTS Reporter node — turned on once the customer's turn starts, and turned off as soon as SR or VA Sync activity is observed again.
Outputs
Audio
none
Events
| name | description |
|---|---|
| VA Silence Detected | Raised when the customer has been silent for at least the configured threshold, or immediately when an SR Unrecognizable Speech event arrives. Carries Data.SilenceType ("empty" for the audio-threshold case, or the SR-provided detected content otherwise) and, for the audio-threshold case, the correct StartSampleIndex/EndSampleIndex for the silence span. |
Remarks
Project Structure
The node needs inputs from VAD Elastic and SR Http. The unmodified audio output of Entry node should reach the input of VA Silence Controller. VA Sync and VA TTS Reporter nodes need to be downstream.
Supported flow types
Batch, Stream