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

# 3. Voice

In the Voice section, you choose and customise exactly how your agent sounds to the customer. It has a few major parts.

## Voice Provider

There are 4 main providers. The rupee figures below are rough cost guides.

| **Provider**                    | **Best for**                                                                                                                                                                 |
| ------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **ElevenLabs** **(\~6 rupees)** | Hindi, English, and Tamil agents. Supports the Breaktime feature, which lets the agent pause between sentences (like a 0.3 second pause after commas) to sound more natural. |
| **Cartesia** **(\~4 rupees)**   | All languages other than Hindi, English, and Tamil (since ElevenLabs is best for those). Breaktime is also available here.                                                   |
| **Sarvam** **(\~4 rupees)**     | All languages, but the Breaktime feature lags a bit.                                                                                                                         |
| **Azure** **(\~2 rupees)**      | Budget bots.                                                                                                                                                                 |

<Warning>
  **Note:** Avoid using emojis and special characters in the system prompt — they can break the agent's behaviour during a live call.
</Warning>

## Voice Models

Every provider has its own voice models. A voice model's job is to turn text into speech (this is called TTS, or Text To Speech).

## Voice

This is the actual voice the customer hears on the call. Every provider offers different voices, and each voice has its own character — some are sharp, energetic, and engaging, while others are calm, soothing, and friendly.

## Voice Parameters

Each provider also gives you settings to fine-tune the voice:

* **Voice Speed** (ElevenLabs, Cartesia, Sarvam, Azure): controls how fast or slow the agent talks.
* **Voice Stability** (ElevenLabs): decides whether the agent sounds like a real human or a robot.
* **Voice Similarity Boost** (ElevenLabs): how closely the agent's voice matches the original voice it was trained on.
* **Voice Style** (ElevenLabs): makes the voice more expressive and dramatic — like turning up the emotion dial.
* **Voice Volume** (Cartesia, Azure): controls how loud or soft the agent speaks.
* **Voice Pitch** (Sarvam, Azure): makes the voice younger and brighter, or deeper and more mature.