Text to speech
Turn writing into a spoken audio file you can download. Slow, and it runs on your device.
- Runs in your browser — nothing you enter is uploaded
- No sign-up, no quotas, free to use as often as you like
How it works
Most text-to-speech sites either read the text aloud without letting you keep the audio, or send your text to a server to get a file back. This generates real audio on your own machine, which you can download — and the text never goes anywhere.
It is slow, and here is why
About 215 MB of model downloads the first time: the speech model and the vocoder that turns its output into sound. Your browser caches both. Generation then runs on your processor and takes several seconds for a sentence, longer for a paragraph. Input is capped at around 600 characters so a single run stays reasonable — split longer text and join the results with the audio joiner here.
The voice
One voice, from the standard research set published with this model. It is clear and natural enough for narration, notes and accessibility, though not indistinguishable from a person. There is no control over emphasis or pace, and unusual names and abbreviations are often mispronounced. Only one speaker embedding is published openly for this model, which is why there is no voice picker.
What you get
A WAV file, which is uncompressed and therefore large. Run it through the audio compressor here for an MP3 if you need something to send or upload.
Why on-device matters
People paste scripts, letters, medical notes and drafts into text-to-speech tools. That text is exactly the kind of thing worth keeping off other people's servers, and here it never leaves the page.
Text to speech — questions
Can I download the audio?
Yes, as a WAV file. Convert it to MP3 with the audio compressor if you need it smaller.
Why is the first run slow?
Around 215 MB of model is fetched once and cached. Generation itself then takes a few seconds per sentence.
Can I choose a different voice?
No. Only one speaker embedding is published openly for this model, so there is a single voice.
Can I control the emphasis or speed?
No. The model takes plain text and produces a steady reading voice.
Is my text uploaded?
No. Everything happens in your browser.