> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zeroruntime.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Turn Detection and Interruptions

> Control when your agent responds, when callers can interrupt it, and how speech is detected.

These settings decide how natural the conversation feels: how quickly the agent replies, and how it handles a caller who talks over it.

<Frame>
  <img src="https://cdn.zeroruntime.ai/zrt/docs-resource/agent-builder/turn-detection.png" alt="Pipeline tab showing the Turn Detection, Response Timing, Interruption Handling, and VAD settings" />
</Frame>

<Note>
  These sections are hidden when you use a realtime (audio) LLM, because the model handles turn-taking itself.
</Note>

## Turn Detection

Detects when the caller starts and stops speaking, so the agent knows when to respond. This setting is per language.

* **Turn Detection Model**: the model that predicts the end of a turn, for example **Echo Large**.
* **Speech Language**: shown when the model supports multiple languages.

Expand **Response Timing** to tune how long the agent waits:

| Setting | Range | Default | Effect |
| - | - | - | - |
| **Minimum Wait Time** | 0–2000 ms | 200 ms | The shortest pause before the agent may reply. |
| **Maximum Wait Time** | 0–5000 ms | 500 ms | The longest the agent waits before treating the turn as finished. |

<Tip>
  If the agent cuts callers off mid-thought, raise **Maximum Wait Time**. If it feels slow to reply, lower **Minimum Wait Time**.
</Tip>

## Interruption Handling

Controls when the caller can interrupt the agent while it is speaking. This setting applies to all languages.

| Mode | Behavior |
| - | - |
| **Hybrid** (recommended) | Adjusts interruption timing dynamically. This is the default. |
| **VAD Only** | Interrupts as soon as speech is detected. Balanced responsiveness. Turns on VAD automatically. |
| **STT Only** | Interrupts once transcribed words arrive. Predictable timing. |

| Setting | Range | Default |
| - | - | - |
| **Minimum Interruption Duration** | 0–1000 ms | 200 ms |
| **Minimum Words to Interrupt** | 0–10 | 1 |

Raise both values to stop background noise or short words like "okay" from interrupting the agent.

## Voice Activity Detection (VAD)

Detects speech and silence in the caller's audio. This setting applies to all languages.

| Setting | Range | Default | Effect |
| - | - | - | - |
| **Speech Sensitivity** | 0–1 | 0.50 | Higher values detect quieter speech, but noise may also count as speech. |
| **Minimum Speech Duration** | 0–1 s | 0.05 s | Sounds shorter than this are ignored. |
| **Silence Detection Delay** | 0–3 s | 0.40 s | How long silence must last before speech is considered stopped. |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.