Phonon Text to Speech
Natural English speech generated entirely in your browser by Phonon, a tiny AI voice model from Gradium. Your text is never uploaded, there is nothing to install, and it costs nothing to use.
The voice that ships with the model.
The voice engine has not been loaded yet.
Works best in a current Chrome, Edge, Firefox or Safari. The engine needs a browser with WebAssembly SIMD.
What is Phonon?
Phonon is a family of small text-to-speech models released as open source by Gradium, a voice AI company that grew out of the Kyutai research lab. The English model on this page has 31 million parameters — tiny next to the cloud models — because it was built for one job: reading English text aloud, on ordinary hardware, without a server. It builds on Kyutai's Pocket TTS design and the Mimi neural audio codec, and it generates speech a fraction of a second at a time, so playback starts almost immediately.
Most free TTS pages send your text to a server, count your characters and wait for you to hit a paywall. This one cannot do any of that, because there is no server in the loop: after a one-time download, the model lives in your browser cache and runs on your CPU. That is also why it keeps working offline.
Private by design
Generation happens on your device. Drafts, scripts and documents you paste here are never transmitted anywhere.
37 voices included
American and British voices plus playful character voices, all from the official open catalogue. Each loads on demand in a fraction of a second.
Reads numbers properly
Before speaking, the engine rewrites figures, dates, prices, emails and web addresses the way a person would say them — the details most cheap voices get wrong.
Good to know
A small on-device model has honest limits. It speaks English only, and it reads best with ordinary written text — articles, scripts, narration, messages. It is a reader, not an actor: for dramatic performances, many languages or a clone of a specific person's voice, a large cloud service is still the right tool. And the first visit needs that one-time download of about 66 MB, which is the price of never paying or uploading anything afterwards.
Questions
Is Phonon text to speech really free?
Yes. The model runs on your own device, so there is no server bill for anyone to pass on to you. There is no account, no usage counter and no premium tier on this page. Use it as much as you like.
Is my text uploaded anywhere?
No. The voice model is downloaded to your browser once and then runs locally, the same way a game runs on your machine. Your text never leaves the tab. You can check this yourself: after the first load, turn off your network and the page still speaks.
Why does the first visit take a moment to load?
The voice engine and model are about 66 MB in total. Your browser downloads them once and keeps them in its cache, so later visits start in about a second and the page even works offline.
Which languages does it speak?
English only, honestly. Phonon also exists for French, German, Spanish and Portuguese, but those models are released under a non-commercial licence, so this page cannot offer them. Feeding it other languages will not produce sensible speech.
Can I use the audio commercially?
The English model on this page is released by Gradium under the Apache 2.0 licence, which allows commercial use, and the voices come from the official Apache-2.0 voice catalogue. As with any tool, you are responsible for the text you feed it and for following the rules of wherever you publish the audio.
How is this different from ElevenLabs or other online voices?
Those are larger cloud models: more languages, stronger acting, voice cloning, and a bill based on how much you generate. This page is the opposite trade: a small model that lives in your browser, free and private, with a single job — reading English text aloud naturally.