> For the complete documentation index, see [llms.txt](https://docs.convai.com/api-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.convai.com/api-docs/plugins-and-integrations/convai-unity-sdk/core-concepts/voice-activity-detection.md).

# Voice activity detection

Voice activity detection (VAD) is how Convai decides whether the audio arriving from a user's microphone contains speech. The Convai Unity SDK exposes `UserVadSettings` so a scene can tune that decision — confidence, timing, and volume thresholds — instead of always relying on Convai's built-in defaults. The SDK resolves these settings once, at the moment a session connects, and sends them as part of the connect request.

`UserVadSettings` applies to Convai's discrete LLM/STT processing path. It does not affect a realtime provider's own voice activity detection or how push-to-talk decides when a turn ends — those paths use their own turn-ending logic, covered in [Turn-taking modes](/api-docs/plugins-and-integrations/convai-unity-sdk/core-concepts/turn-taking-modes.md).

***

### Why voice activity detection is a connect-time setting

`UserVadSettings` tunes Convai's audio pipeline rather than a value a scene needs to change moment-to-moment. A factory-floor training simulation with background noise needs a higher `MinVolume` and `Confidence` threshold than a quiet home-office scene, but neither scenario needs to change that threshold while a conversation is already in progress. The SDK reflects this: it reads `UserVadSettings` once, immediately before it builds the connect request for a session, and does not expose any API to push updated values into a session that is already `Connected`.

This also means `UserVadSettings` behaves differently from `TurnTakingOptions`. `RoomSessionConnectOptions` — the per-invocation override object passed to a connect call — carries a `TurnTaking` field, so turn-taking behavior can be overridden per connection attempt. It has no equivalent field for `UserVadSettings`. The only sources the SDK reads from are the `ConvaiRoomManagerProfile` asset or the equivalent inline field on `ConvaiRoomManager` — there is no per-connect override.

***

### `UserVadSettings` field reference

| Field              | Type    | Default | Valid range   | Description                                                                                                                                                            |
| ------------------ | ------- | ------- | ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `UseServerDefault` | `bool`  | `true`  | —             | When enabled, the SDK omits `vad_params` from the connect request entirely and Convai applies its own defaults. Disable it to send the four fields below instead.      |
| `Confidence`       | `float` | `0.7`   | `0`–`1`       | Confidence threshold Convai's voice activity detector requires before treating incoming audio as speech.                                                               |
| `StartSecs`        | `float` | `0.2`   | `≥ 0` seconds | Duration of continuous speech required before Convai treats the user's speech as started.                                                                              |
| `StopSecs`         | `float` | `0.2`   | `≥ 0` seconds | Duration of silence required before Convai treats the user's speech as stopped. Convai may clamp this to `0.2` seconds when hands-free smart-turn detection is active. |
| `MinVolume`        | `float` | `0.6`   | `0`–`1`       | Minimum input volume Convai's voice activity detector requires before considering audio a speech candidate.                                                            |

The four numeric defaults mirror the defaults Convai itself applies server-side, so leaving `UseServerDefault` enabled and omitting `vad_params` produces the same detection behavior as sending these exact values explicitly.

***

### Server default vs. custom values at connect

Each time a session connects, the SDK resolves `UserVadSettings` from its configured source and decides what to send.

```mermaid
graph TD
    A["Session connect requested"] --> B["Resolve UserVadSettings from ConvaiRoomManagerProfile or scene defaults"]
    B --> C{"UseServerDefault?"}
    C -->|"true (default)"| D["Omit vad_params from the connect request"]
    C -->|"false"| E["Build vad_params from Confidence, StartSecs, StopSecs, MinVolume"]
    D --> F["Convai applies its own defaults"]
    E --> G["Convai applies your custom values"]
```

When `UseServerDefault` is `true`, the connect request's `vad_params` field is left `null`, and the SDK's request serializer omits `null` fields entirely — `vad_params` does not appear in the payload sent to Convai at all. When `UseServerDefault` is `false`, the SDK sends `vad_params` as an object with `confidence`, `start_secs`, `stop_secs`, and `min_volume`.

The SDK logs the resolved decision on every connect attempt:

* Server default: `Using server default vad_params (field omitted from connect payload).`
* Custom values: `Sending custom vad_params: confidence=<value>, start_secs=<value>, stop_secs=<value>, min_volume=<value>`

{% hint style="info" %}
These log lines come from the SDK's internal logger and only appear if your project's log level surfaces `Debug`- and `Info`-severity messages.
{% endhint %}

***

### Author custom values with `ConvaiRoomManagerProfile`

`ConvaiRoomManagerProfile` is a `ScriptableObject` asset that stores a serialized `UserVadSettings` field, defaulting to `UseServerDefault = true`. Its public `UserVadSettings` property returns a clone of that field, so reading it never lets a caller mutate the asset's stored values by reference.

To send custom detection thresholds instead of Convai's defaults, open the `ConvaiRoomManagerProfile` asset in the Inspector, disable `Use Server Default`, and set `Confidence`, `Start Secs`, `Stop Secs`, and `Min Volume` to the values your scene needs. Any `ConvaiRoomManager` that references this profile as its configuration source sends those values on its next connect. The same fields also exist inline on `ConvaiRoomManager` itself, for scenes that configure room settings directly on the component instead of through a shared profile asset.

{% hint style="warning" %}
`UserVadSettings` has no public setter and no per-connect override on `RoomSessionConnectOptions`. Set it on the `ConvaiRoomManagerProfile` asset or the inline field on `ConvaiRoomManager` before the scene connects — changes made while a session is already `Connected` do not apply until the next connect attempt.
{% endhint %}

***

### Next steps

You now know what `UserVadSettings` controls, how the SDK decides between Convai's defaults and your custom values, and where to configure it. Read Turn-taking modes to see how voice activity detection relates to hands-free turn-ending behavior, then Event system to subscribe to session and character events at runtime.

{% content-ref url="/pages/b9e67d1d27aaccf9e9bc9b260045411047aa37c3" %}
[Turn-taking modes](/api-docs/plugins-and-integrations/convai-unity-sdk/core-concepts/turn-taking-modes.md)
{% endcontent-ref %}

{% content-ref url="/pages/bcd284a409f00820e1693ab998dc7a99a4fbd780" %}
[Event system](/api-docs/plugins-and-integrations/convai-unity-sdk/core-concepts/event-system.md)
{% endcontent-ref %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.convai.com/api-docs/plugins-and-integrations/convai-unity-sdk/core-concepts/voice-activity-detection.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.