GUIDE

An offline alternative to OpenAI text to speech

OpenAI’s speech endpoint made text to speech a single API call. According to OpenAI’s documentation, its models are now deprecated and shut down on January 6, 2027, and the replacement works through a different API. If the integration has to be rewritten anyway, it is a good moment to ask whether speech should leave your servers at all.

The free SuperClonica engine runs Supertonic 3 as your own HTTP service, on an ordinary processor, with stock, catalog or trained voices. This guide compares the two honestly and shows how to swap the endpoint.

Browse voices Install the SuperClonica engine

Why switch now

OpenAI’s voices are good and quick to start with; the reasons to switch are control and cost, not quality.

  • The speech endpoint’s models, tts-1, tts-1-hd, gpt-4o-mini-tts, are deprecated and shut down on January 6, 2027.
  • Their replacement works through the Realtime API, a different workflow, so the code changes anyway.
  • Every request sends your text to a cloud service and is billed by usage.
  • Without internet access and an API key, there is no speech.

Compare the sound free

On the page of any voice, paste a typical sentence from your app, 30 to 100 characters, choose the language and press Voice it. In a few seconds you hear whether it can replace the voice you use now, with no account.

Swap the endpoint

Most applications call speech in one place, so the change is usually a few lines.

  • Start the server on any processor: docker run -d -p 8000:8000 -v superclonica-models:/models superclonica/superclonica:cpu.
  • Send POST /synthesize with text, voice and lang; the answer is a 16-bit mono WAV file.
  • The format is not OpenAI-compatible: a small adapter maps input to text and your voice names to M1–M5, F1–F5 or your own voice files.
  • Set SUPERCLONICA_API_KEY and keep the server behind HTTPS or on a private network.

Your own voices

Instead of a fixed set of built-in voices, choose from more than 170 catalog voices, each with a free demo, or from the free voices in the superclonica/supertonic-voices repository on GitHub. For a voice of your own, train it in the SuperClonica studio from a clean recording of a speaker who agreed, on a computer with an NVIDIA graphics card, or order it on the exchange. The voice file travels inside each request, so the server stores no voices.

When OpenAI stays the better choice

Supertonic 3 speaks 31 languages, and Chinese is not among them; OpenAI’s speech covers Chinese and more. The engine takes no written instructions about tone: delivery comes from the voice and the punctuation. And a hosted API needs no server of your own. If these matter most, stay with OpenAI; if privacy, cost and your own voices matter more, host speech yourself.