# How it works: voices, local speech, training

How to choose a voice, run speech locally on your CPU with Supertonic 3, train your own voices and sell them in the SuperClonica marketplace.

Fast local TTS for your projects and marketplace for your voices.

## Good voice. Minimum resources.

SuperClonica — custom voices for Supertonic 3. The engine turns text into speech locally on the CPU. You can find a voice for your product or create your own and put it up for sale.

### On desktop

Turn text into speech locally on your CPU. No dedicated GPU required.

### In the browser

Build voice into interfaces with ONNX Runtime Web.

### On mobile

Add local speech to compatible mobile apps.

### In your product

Add speech synthesis to your own environment with the local engine.

## How to use a downloaded voice

The JSON file contains the voice profile. To generate speech, you also need the Supertonic 3 engine and its base model.

### 1. Download the JSON file

After acquiring a voice, open “My library” and download its JSON file.

### 2. Install the engine

Install the SuperClonica engine for Supertonic 3 with one pip command. The base model downloads on first run; after that, speech is generated locally.

### 3. Turn text into speech

Load the JSON file in a compatible client, choose a language and enter your text. A Python example is provided below.

## Create a voice that people want to buy

### Why create your own voice?

The open release of Supertonic 3 includes ready-made presets. It has no custom voice matching your reference recording — you need to create one separately.

### Voice matching software

We provide our own software for matching a custom voice to a reference.

### Neural network for fast voice matching

Our neural network, trained on hundreds of ready-made custom voices, helps you match the sound to your reference faster.

### 1. Create a voice

Give your voice a name. Add reference recordings and manage their training runs on its page. You can change the name and language later.

### 2. Add a reference

Upload a clear recording 5 to 10 seconds long and enter its exact transcript. Each reference has its own training history. Recordings stay private.

### 3. Train and listen

Connect your worker. Click “Train” beside a reference recording, choose the step count and start training. Compare checkpoint demos and similarity scores, then continue from a result you like.

### 4. Save or publish

Listen to the checkpoints and download the JSON for free from the training page. To sell the model, open “Listings”, confirm it is ready and add a cover, description and price.

### 5. Start earning

Set your prices in US dollars. Proceeds after the platform fee are credited to your balance. Withdrawals are not available yet.

### 6. Stay in control

Your voices, files and queue are private. The categories are set by the administrator. Buyers have access to the purchased version of the model, and the references remain with the author.

### Rights and usage

Only publish models based on recordings you have the rights to and the voice owner’s consent for. A purchased model is intended for use in your own projects; reselling or republishing someone else’s model is not allowed.

- [SuperClonica](https://superclonica.com/en/)
- [AI voice catalog for text to speech](https://superclonica.com/en/explore)
- [Custom voice commissions for Supertonic 3](https://superclonica.com/en/requests)
- [How it works: voices, local speech, training](https://superclonica.com/en/guide)
- [Install the faster Supertonic 3 engine](https://superclonica.com/en/superclonica)
- [About us: speech synthesis enthusiasts](https://superclonica.com/en/about)
- [Contacts, support and voice complaints](https://superclonica.com/en/contacts)
- [Privacy policy: what data we keep](https://superclonica.com/en/privacy)
- [Terms of use: rules for buyers and authors](https://superclonica.com/en/terms)
- [Text to speech](https://superclonica.com/en/text-to-speech)