Control Turn Taking with VAD Modes

Turn taking decides when the caller has finished speaking and the agent may reply. The turnTaking feature selects the voice activity detection (VAD) mode for the main customer call. What you’ll learn: What each VAD mode does, how to select one, and how the Flux mode falls back on unsupported languages.

Configuration

Set the mode at config.features.turnTaking:

{
  "config": {
    "features": {
      "turnTaking": {
        "version": "v1",
        "vad": "asap_v2"
      }
    }
  }
}

In the UI, the setting lives on the Voice & Speech tab under Turn Taking.

VAD modes

Mode	Behavior
`asap_v1`	Legacy VAD behavior.
`asap_v2`	Improved VAD tuned for faster turn detection. Default.
`flux`	Runs speech recognition and turn detection on the Deepgram Flux pipeline, which models end of turn directly instead of relying on silence length.

Omitting turnTaking or vad uses asap_v2.

Flux language support and fallback

The Flux pipeline supports these languages: English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch.

If the agent’s language is not in this list, or the Flux connection fails at call start, the call automatically falls back to the legacy VAD pipeline. The call proceeds normally either way.

Scope

The setting applies only to the main customer call. Warm-transfer and operator legs keep their own defaults.

⌘I

Introduction

Create

Test

Deploy

Monitor

Advanced Features

Troubleshooting

Configuration

VAD modes

Flux language support and fallback

Scope

​Configuration

​VAD modes

​Flux language support and fallback

​Scope

Configuration

VAD modes

Flux language support and fallback

Scope