Skip to main content

Voice Activity Detection

Player voice activity detection (VAD) is a feature of Inworld's Unreal Engine SDK that allows a game to detect when a player begins speaking, which can then be used trigger in-game events.

VAD Options​

To enable VAD, select a speech mode under Speech Options in the Inworld Session Component.

vad-options

Speech Modes​

There are three Speech Modes available:

  • Default: VAD is disabled, all captured audio data is sent during audio sessions, VAD blueprint events do not trigger.
  • VAD Continious Audio Send: VAD is enabled, all captured audio data is sent during audio sessions, VAD blueprint events trigger.
  • VAD Conditional Audio Send: VAD is enabled, only data with detected player voice is sent during audio sessions, VAD blueprint events work.

Speech Options​

In addition, three further Speech Options are available:

  • VAD Prob Threshold: The probability threshold the language processing model uses to determine if VAD should be triggered.
  • VAD Buffer Chunks Num: The number of 100 ms audio chunks that are buffered to send to server upon voice detection.
  • VAD Silence Chunks Num: The number of 100 ms audio chunks of silence required before VAD is considered to have detected silence.

Triggering VAD Events​

VAD blueprint events can be triggered from either the Inworld Player Component or Inworld Character Component.

vad-options vad-options