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

# Voice Engines

> Pick the voice engine a Standard campaign speaks with, and what happens when the voice service fails mid-call.

On a [Standard](https://docs.topcalls.ai/guides/live-mode.md) campaign, the voice engine turns the agent's words into speech. Live campaigns don't use one: their voice model hears and speaks by itself.

## Where to pick it

**1. Open How it sounds**

In the campaign editor, open the **How it sounds** step. Under **Engine**, **How it works** must be **Standard**.

**2. Choose Custom**

Under **Setup**, select **Custom**. **Recommended** keeps the setup that fits most campaigns, and you can stop there.

**3. Pick the voice engine**

Open **Voice engine** and pick one. Check that it speaks the language set under **Callers speak**: the table below and `GET /v1/models` list each engine's languages.

**4. Tune and listen**

The sliders under **Advanced** change with the engine you picked. Play the voice preview, then place a [test call](https://docs.topcalls.ai/guides/test-your-campaign.md).

## Which engine to pick

The recommended engine fits most campaigns. Change it when the campaign needs one of these:

| Pick | When |
| - | - |
| **Flash v2.5** (recommended) | Fast replies in 32 languages. The default for every Standard campaign. |
| **Turbo v2.5** | A little more quality, a little slower, the same 32 languages. |
| **Multilingual v2** | The most lifelike reading in 29 languages. The only engine with the **Voice expressiveness** slider. |
| **v3 Conversational** | The most expressive delivery, at a fixed speaking pace. It has no sliders. |
| **v4 Turbo** | Fast replies and the longest language list of any engine. It has one slider, **Voice consistency**. |

`GET /v1/models` lists the same engines under `legacy.tts`, each with its `supported_languages` and the settings it accepts. Send the row's `provider` and `id` as `tts_provider` and `tts_model`.

The languages a campaign can use depend on both speech recognition and the voice engine. A language counts only when both handle it.

## Cloned voices

Every engine in the list can speak a cloned voice. Clones run on Standard campaigns only. See the [Voice Library](https://docs.topcalls.ai/guides/custom-voices.md).

## When the voice service fails mid-call

On a Standard call, if the voice service stops working in the middle of the call, the call switches to a backup voice engine. The lines the failure cut off are sent again on the backup, and the call carries on in the backup voice until it ends. Each built-in voice has its own paired backup voice. A cloned voice falls back to a standard backup voice. Either way the caller hears a different voice, but the conversation doesn't stop.

The backup is used only when it speaks every language the call uses. When it doesn't, or when the backup fails too, the call ends with the reason `tts_failed` instead of leaving the line silent. You don't set anything up for this, and you can't pick the backup engine for a campaign.

## Settings you don't need anymore

`tts_emotion`, `tts_volume` and `emotion_enabled` are accepted and ignored. The engine that used them is no longer offered, and campaigns that used it were moved onto the engines in this list.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.