> This page is for Voice APIs.

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://dev.hume.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://dev.hume.ai/_mcp/server.

# Turn detection

**Turn detection controls how EVI picks up speech and determines when a user has finished their turn.** Configure
these settings through the optional `turn_detection` field on an EVI config. All fields fall back to defaults
when omitted.

Tuning turn detection lets you shape the pacing of a conversation. You can make EVI wait longer before responding,
react more quickly when the user finishes speaking, or filter out background noise more aggressively depending on
your application's needs.

#### [API Reference](/reference/speech-to-speech-evi/configs/create-config#request.body.turn_detection)

See the full API Reference for turn detection.

## Supported parameters

### 1. End of turn silence

`end_of_turn_silence_ms`: How long the user must be silent before EVI considers their turn complete and begins
generating a response.

* Lower values lead to faster responses but increase the chance of EVI responding during a mid-thought pause.
* Higher values give users more room to pause without triggering a response.

**Default**: 800ms | **Range**: 500ms to 3,000ms

### 2. Speech detection threshold

`speech_detection_threshold`: How confident the system must be that audio contains speech before it begins processing.

* Lower values increase sensitivity, capturing softer speech at the cost of more noise-triggered processing.
* Higher values require clearer audio to register as speech, reducing false activations from background noise but
  potentially missing quieter speech.

**Default**: 0.5 | **Range**: 0.0 to 1.0

### 3. Prefix padding

`prefix_padding_ms`: The duration of audio captured before the detected start of speech. This ensures the beginning of
an utterance is not clipped.

* Higher values preserve more of the lead-in to speech, which can improve transcription accuracy for words that
  start abruptly.

**Default**: 300ms | **Range**: 0ms to 1,000ms

## Example configurations

### Patient turn-taking

Give users more time to pause mid-thought without EVI treating the silence as a completed turn. This is useful when
users need to think before finishing a sentence, such as in language learning or interview preparation.

```json
{
  "turn_detection": {
    "end_of_turn_silence_ms": 2000
  }
}
```

> **Tip**
>
> For a fully patient interaction, pair this with a higher `min_interruption_ms` in the
> [interruption](/docs/speech-to-speech-evi/configuration/interruption) config so hesitations don't cut off EVI
> mid-response.

### Responsive turn-taking

Make EVI begin responding sooner after the user stops speaking. This creates a snappier conversational feel suited to
fast-paced interactions like scheduling or quick Q\&A.

```json
{
  "turn_detection": {
    "end_of_turn_silence_ms": 500
  }
}
```

### Noise-tolerant detection

Require clearer audio before treating input as speech and capture more lead-in audio to prevent clipping in noisy
conditions.

```json
{
  "turn_detection": {
    "speech_detection_threshold": 0.7,
    "prefix_padding_ms": 500
  }
}
```

---