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

# Choosing a TTS voice

> Local text-to-speech and voice cloning for ESP32 AI toys with Qwen3-TTS, Chatterbox, Breeze, OmniVoice, Voxtral, Higgs, Fish Audio, rumik and Irodori on Apple Silicon.

OpenToys ships ten TTS backends. Pick one on the **Models** page and swap it live. Qwen3-TTS is the default and the best fit for most Apple Silicon Macs.

| Backend | Best for | RAM | Page |
| - | - | - | - |
| **Qwen3-TTS** (default) | Balanced, most Macs | 4 GB | [Qwen3-TTS](https://elatoai.com/docs/local/tts/qwen3-tts.md) |
| **Chatterbox-turbo** | Expressive, cloning | 6 GB | [Chatterbox-turbo](https://elatoai.com/docs/local/tts/chatterbox.md) |
| **BreezeBlue TTS 2** | Expressive voice design | 8 GB | [BreezeBlue](https://elatoai.com/docs/local/tts/breeze-blue.md) |
| **Chatterbox Multilingual v3** | 23 languages | 6 GB | [Multilingual](https://elatoai.com/docs/local/tts/chatterbox-multilingual.md) |
| **OmniVoice** | 646+ languages, small and fast | 4 GB | [OmniVoice](https://elatoai.com/docs/local/tts/omnivoice.md) |
| **Voxtral TTS** | Natural, 20 built-in voices | 5 GB | [Voxtral](https://elatoai.com/docs/local/tts/voxtral.md) |
| **Higgs Audio v3** | Conversational, 100 languages | 12 GB | [Higgs Audio](https://elatoai.com/docs/local/tts/higgs-audio.md) |
| **Fish Audio Drama 3** | Studio quality, multi-speaker | 26 GB | [Fish Audio](https://elatoai.com/docs/local/tts/fish-audio.md) |
| **rumik-oss 1** | Indic languages | 3 GB | [rumik-oss](https://elatoai.com/docs/local/tts/rumik.md) |
| **Irodori-TTS** | Japanese | 3 GB | [Irodori](https://elatoai.com/docs/local/tts/irodori.md) |

## Choosing a backend

Check the licenses before shipping a product. Voxtral, Higgs Audio and rumik-oss are non-commercial or research-only.

## Cloning a voice

Most backends clone a voice from less than 10 seconds of audio. Voxtral and rumik-oss use built-in voices instead.

Add a voice with a short, clean reference clip and its transcript, then assign it to a character. If a character's voice isn't on disk yet, the Play button shows **Download voice**.

> [!NOTE]
>
> Want cloud voice cloning instead? See [Eleven Labs](https://elatoai.com/docs/models/eleven-labs.md) and [Hume](https://elatoai.com/docs/models/hume.md).


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