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

# Voice & Speech

> Configure text-to-speech and speech-to-text settings

Configure how your agent sounds — the voice, speaking speed, and speech recognition.

## Text-to-Speech (TTS)

The TTS provider converts the agent's text responses into spoken audio.

### Providers

| Provider            | Voices             | Notes                                       |
| ------------------- | ------------------ | ------------------------------------------- |
| **Cartesia**        | 10+ voices         | Default. High quality, fast.                |
| **ElevenLabs**      | Multiple           | Premium voice quality, custom voice cloning |
| **Gemini Live**     | Puck, Fenrir, etc. | Only with Gemini Live LLM                   |
| **OpenAI Realtime** | marin, etc.        | Only with OpenAI Realtime LLM               |

### Voice selection

Each provider offers a set of predefined voices. Select from the dropdown in the agent detail page.

**Custom voice:** If you have a custom voice model (e.g., a cloned voice on ElevenLabs), enter the voice UUID in the custom voice ID field.

### Speaking rate

Adjust how fast the agent speaks:

| Rate    | Effect                 |
| ------- | ---------------------- |
| **0.5** | Very slow              |
| **1.0** | Normal speed (default) |
| **1.5** | Noticeably faster      |
| **2.0** | Maximum speed          |

**Note:** Speaking rate is not adjustable for Native Realtime providers (Gemini Live, OpenAI Realtime).

## Speech-to-Text (STT)

The STT provider transcribes the caller's speech into text for the LLM.

| Provider       | Notes                                              |
| -------------- | -------------------------------------------------- |
| **Deepgram**   | Default. Fast and accurate for German and English. |
| **ElevenLabs** | Good multilingual support                          |
| **Mistral**    | Alternative option                                 |

**Note:** STT is not configurable for Native Realtime LLMs — they have built-in speech recognition.

## Who can edit

| Role                    | Voice/TTS | Speaking Rate |   STT  |
| ----------------------- | :-------: | :-----------: | :----: |
| Super-Admin / Dev-Admin |    Full   |      Full     |  Full  |
| Client-Admin            |    Full   |      Full     | Hidden |
| Client-Employee         | Read-only |   Read-only   | Hidden |