Skip to content
CUSTOM VOICES · SUPERTONIC 3

Fast text to speech.
Your sound.

Custom voices for Supertonic 3. Generate speech on your CPU, on desktop, in the browser or in mobile apps. Find a ready-made voice, or create and sell your own.

Find a voiceCreate a voice for free
New SuperClonica engine: up to +73% speed
Runs on your CPUOffline speech synthesisFree training

SuperClonica — Text to speech in many AI voices online

Text to speech in many voices: game and cartoon characters, celebrities, announcers. Try any voice free online, then generate speech offline on your computer.

High quality. At any scale.

Natural speech synthesis

Natural, expressive sound for dialogue, games, videos and long stories.

For any project

Embed it in small apps with minimal resources. The compact model runs on a CPU, no dedicated GPU required.

For large platforms

For services with a large flow of speech. Run the model on your own servers and add capacity as your audience grows.

SuperClonica. Faster, cleaner, steadier.

Our engine built on Supertonic 3: the same model and the same voices, with cleaner sound and faster synthesis. Marketplace voices work without changes.

speed on CPU

Measured on a Ryzen 7 5800X: average compared with standard Supertonic 3.

speed on GPU

Measured on an RTX 5070 Ti: average compared with standard Supertonic 3.

Cleaner and steadier

Silence between words without hiss and clean phrase endings. Long text is spoken in short pieces, so there are fewer repeats and cut-offs.

Compatible with Supertonic 3 JSON

Marketplace voices and your own Supertonic 3 JSON files work without conversion.

Good voice. Fewer resources.

Quality, CPU speed and timbre stability for everyday speech synthesis.

Scroll through the table to compare models →

Stars and speed on the Ryzen 7 7800X3D are SuperClonica's rough estimates, not measured results.

Generation time = audio length × RTF. Speed relative to real time = 1 / RTF. Ranges are a forecast for 8 cores / 16 threads and DDR5, accounting for different CPU clients. The conversion factors below are our assumptions and still need to be verified on a 7800X3D. Warmed-up model, single request, weight loading excluded. Speed depends on memory, client and settings.

SuperClonica. On average 61% faster than Supertonic 3 on CPU (measured on a Ryzen 7 5800X) and 73% faster on GPU (RTX 5070 Ti) with the same settings, so ≈6.4–9.7× real time.

Supertonic: RTF 0.20 on CPU with 16 threads. We allow for a deviation of about 20% in speed: 4–6× realtime. The authors did not indicate the CPU model.

Fish: estimate based on the GGUF client and Dual-AR with 10 codebooks. We assume an effective memory bandwidth of 30–50 GB/s plus compute and codec overhead: RTF 4–10. This is the roughest estimate, with no CPU benchmark.

Qwen: 0.6B/INT4 on Ryzen 6800H: RTF 2.02. We assume a speedup of roughly 1–2×: RTF 1–2. We estimate version 1.7B separately, with a wider RTF range of 2–5. Plain PyTorch can be significantly slower.

OmniVoice: Client CPU report: RTF 4.92 at 32 steps. We assume a speedup of up to 2.5× relative to this configuration: RTF 2–5. The exact CPU model is not specified in the report.

VoxCPM 2: INT8 on Core i3-12300: RTF 3.8. With an assumed speedup of 1–2×, we get a rounded RTF range of 2–4.

F5: RTF 37.284 on a single thread Xeon 8457C, 32 steps. With an assumed effective speedup of 2.5–7.5× from the CPU and parallelism, we get RTF 5–15. We do not assume linear scaling to 16 threads.

Try any voice. Free.

Voice text in any voice

On the page of every voice in the marketplace you can voice your own phrase of up to 100 characters for free and download the recording.

Free voices

Some voices in the catalog are free: download their JSON without signing up and use them in Supertonic 3 or SuperClonica.

Find a voice that fits your project.

Need a voice based on your recording?

Describe the sound and set a budget. An author will train a voice for your task.

One JSON. Across your devices.

On desktop

In the browser

On mobile

In your product

Download the JSON and use it with Supertonic 3 or our SuperClonica engine.

Train your voice. Free.

Train a voice on your own computer or server. Download the Supertonic 3 JSON to use in your projects, or sell the model.

Voice from your reference

Supertonic 3 comes with ready-made presets. Create your own timbre from your recording.

Matching software

Our software helps you match a voice to a reference recording and manage training in Creation.

Faster voice matching

Our neural network is trained on hundreds of custom voices and helps you quickly find the right sound.

Sell voices and fulfill orders

Set your prices in dollars. Sell ready-made voices or train custom models from customers’ reference recordings.

Manage training and workers in Creation. Find listings, sales and orders in Monetization.

Earnings example

Voices stay with you. No subscriptions, no cloud.

Speech on your own computer

Synthesis runs locally: no cloud, no subscription, no character limits. Your texts never leave your machine.

Faster than Supertonic 3

The SuperClonica engine is on average 61% faster on the CPU and 73% faster on the GPU, with the same voices.

Buy once, keep forever

Pay for a voice once. Its JSON stays in your library, even if the author updates or removes the voice.

31 languages in every voice

One voice speaks 31 languages — you choose the language when you synthesise.

Try before you buy

On a voice’s page you can voice your own phrase and hear demos in 8 languages. There are free voices too.

Your own voice for free

Train a voice on your own graphics card with SuperClonica Worker — the platform charges nothing for training.

Earn with voices

Sell voices in the catalog and take orders on the voice commissions board. Earnings go to your balance.

An open format

A voice is a plain Supertonic 3 JSON: it also works in the original engine, with no lock-in to our platform.

Frequently asked questions

What is SuperClonica?

It is a voice catalog for speech synthesis and its own SuperClonica engine built on Supertonic 3. Authors train and sell voices; you pick the sound and voice your texts on your own computer.

Can I try a voice for free?

Yes. On any voice’s page you can voice your own phrase of up to 100 characters and hear demos in 8 languages. And free voices download in full — no account needed.

How do I buy a voice?

Prices are in US dollars; a voice is bought once and kept forever.

Top up your balance with cryptocurrency via NOWPayments — no card details needed.

Press “Buy voice” on its page.

Download the JSON from your library. It stays there even if the author updates or removes the voice.

How do I voice text on my own computer?

Install SuperClonica or Supertonic 3 and pass the engine the path to the voice’s JSON. Synthesis runs on an ordinary CPU; a graphics card only speeds it up. Step by step — in “How it works”.

Which devices and languages do the voices work with?

A voice is one JSON file. It works on Windows, macOS and Linux, in the browser via ONNX Runtime Web, in mobile apps, games and bots. Every voice speaks 31 languages.

How do I create my own voice?

You need one clean voice recording and a computer with a graphics card.

Create a voice in the studio and upload a 5–10 second recording of clean speech.

The recording’s language and words are recognised automatically — you only check them.

Press “Train”. You can do it right away, while the recording is still being recognised: the training joins the queue by itself.

What do I need for training?

A computer with an NVIDIA graphics card with 8 GB of memory or more (12–16 GB is better) and the SuperClonica Worker app. Training is free and runs on your hardware; follow it in the queue: progress, time left and the similarity of every checkpoint.

How do I sell my voices?

Prepare the card in Listings: a cover, a description and a price or free access. After a moderator approves it, the voice appears in the catalog, demos in 8 languages are made automatically and the description is translated into every language of the site. Earnings go to your balance.

Can I order a custom voice?

Yes, on the voice commissions board. Upload recordings and describe the sound and the budget. The budget is reserved on your balance and goes to the author only after you accept the result.

Why choose SuperClonica over other voice services?

Voices run locally and are bought for good — no subscriptions, cloud or character limits. All the advantages are in “Why SuperClonica”, and the comparison with other models is on the “Text to speech” page.

Where do I go if something does not work?

Open “Support” in the account menu: describe the problem and attach screenshots or worker logs — we answer right there. You can also write to support@superclonica.com.

Text-to-speech guides

Voice marketplace

Loading SuperClonica…

Reload page