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

# Voice and audio updates

> Change voice, fix pronunciations, adjust latency, and manage cached audio for optimal voice quality.

Update voice settings, fix mispronunciations, manage cached audio, and tune call behavior to maintain high-quality voice experiences for your callers.

## Quick reference

| I need to...             | Action                                                 |
| ------------------------ | ------------------------------------------------------ |
| Change the agent's voice | **Voice > Settings** → Change                          |
| Adjust voice parameters  | **Voice > Settings** → settings gear                   |
| Fix a mispronunciation   | **Voice > Advanced settings > Speech** → Pronunciation |
| Update cached audio      | **Voice > Audio library** → Edit → Sync                |
| Enable/disable barge-in  | **Voice > Advanced settings > Call** → toggle          |
| Upload custom audio      | **Voice > Audio library** → Upload                     |

## Changing your agent's voice

Consider updating when your brand refreshes, customers report clarity issues, you're expanding to new languages, or newer voice models become available.

1. Go to **Voice > Settings**
2. Click **Change** to open the [Voice Library](/agent-studio/voice/voice-design/voice-library)
3. Filter by **Language**, **Region**, and **Gender**
4. Preview voices with custom text
5. Click **Select** to apply
6. Test in Agent Chat before publishing

> **Tip**
>
> For non-English projects, use a `multilingual_v2` model to ensure proper language support.

For programmatic voice configuration, see [voice classes](/agent-studio/tools/classes-and-objects/voice) and [Add a voice](/agent-studio/voice/voice-design/add-a-voice).

## Barge-in

Toggle in **Voice > Advanced settings > Call**. Lets callers interrupt the agent mid-sentence.

**Enable when:** callers frequently interrupt, or you want more natural conversations.
**Disable when:** delivering complete information (legal disclaimers), background noise causes false interruptions.

## Managing audio quality

### Cached audio

The Audio library tab lets you cache and optimize frequently-used audio for reduced latency and consistent quality.

* Open **Voice > Audio library**
* Click **Edit** to adjust stability/clarity settings or add IPA pronunciation corrections
* Click the **sync** icon to regenerate, then preview

> **Tip**
>
> Audio is only cached after the same TTS is generated at least twice in 24 hours. For critical phrases (greetings, transfers), generate them repeatedly or upload manually.

### Custom audio uploads

Upload pre-recorded audio (WAV or MP3) for maximum control over greetings, legal disclaimers, or brand-specific moments.

## Fixing pronunciations

When the agent mispronounces words:

1. Go to **Voice > Advanced settings > Speech** → **Pronunciation** section
2. Add a pronunciation rule
3. Enter the regex pattern for the word as it appears in text
4. Provide the IPA replacement (e.g., "PolyAI" → `/ˈpɒli eɪ aɪ/`)
5. Test in Agent Chat

You can also use SSML for advanced control:

```xml
<break time="500ms"/>
<prosody rate="slow">Speak this slowly</prosody>
```

## Troubleshooting

| Issue                          | Likely cause               | Fix                                                  |
| ------------------------------ | -------------------------- | ---------------------------------------------------- |
| Voice sounds robotic           | Low-quality TTS            | Switch to Cartesia or ElevenLabs                     |
| Agent speaks too fast          | Rate set too high          | Adjust with the settings gear in Voice Settings      |
| Agent interrupts frequently    | Barge-in too sensitive     | Disable barge-in in Advanced settings > Call         |
| Mispronunciations              | TTS doesn't recognize word | Add pronunciation rule in Advanced settings > Speech |
| High latency                   | Slow TTS provider          | Switch to Cartesia or use cached audio               |
| Background noise interruptions | Barge-in too sensitive     | Disable barge-in or increase speech end delay        |

## Maintenance routine

* **Monthly:** Listen to recent calls and identify voice quality issues
* **As needed:** Add pronunciations for new terms
* **After voice changes:** Regenerate cached audio

## Related pages

* [Audio library](/agent-studio/voice/audio-library) – audio caching and optimization
* [Advanced voice settings](/agent-studio/voice/advanced-settings) – model, barge-in, speech recognition, pronunciation
* [Voice library](/agent-studio/voice/voice-design/voice-library) – browse and select voices
* [Voice settings](/agent-studio/voice/settings) – voice configuration options