Text to speech
Type or paste your text, or drop in a PDF, a Word file or a photo of a page. Pick a voice, listen, and download it as an MP3. The speech is generated on your own device — your document never leaves your computer.
Getting a natural-sounding result
The voice reads exactly what is written, and that is a strength once you work with it. It has no idea what your text means, so it takes all its timing from punctuation: a full stop is a pause, a comma a shorter one, and a paragraph without either becomes a breathless run. The single biggest improvement you can make is not a setting — it is breaking long sentences into shorter ones and putting the commas where you would actually breathe.
Prepare the text first
Paste your text into the box and read it once before you press Speak. Take out anything that belongs to the page rather than to the sentence: page numbers, running headers, footnote markers, the leftovers of a PDF layout. Write out anything unusual the way it should sound — an uncommon name, a brand, a foreign word — because the voice will attempt it literally. Numbers, amounts and common abbreviations are already handled, so you can leave those alone.
What people use this for
Proofreading is the use most people discover by accident: hearing your own text read back catches missing words and clumsy sentences that the eye slides straight over. Beyond that — a voice-over for a video without recording anything, a long document turned into something you can listen to while driving, a text made accessible for someone who reads with difficulty, and language practice where you want to hear a sentence rather than read it. Because everything runs on your own device, a contract or a medical letter is safe to paste in.
How it works
A speech model runs inside your browser and turns your writing into a voice — nothing is uploaded, so a contract, a diagnosis or a private letter stays on your machine. A PDF is read straight from its text layer, a Word file from its own contents, and a photo of a page goes through our OCR first, so a book or a printed letter works too. Numbers, amounts and symbols are written out before speaking, in the language you picked, so "€25" is read as "twenty-five euro" rather than garbled. The first time you use it, the voice engine downloads once — about 400 MB on a computer whose graphics card can run it, which then speaks almost instantly, and about 150 MB otherwise — and your browser keeps it after that.
Which languages can it speak?
All ten languages of the site: English, Dutch, German, French, Spanish, Italian, Polish, Portuguese, Turkish and Arabic. Pick the language of your text next to the voice; every voice speaks every language, and numbers, dates, amounts and clock times are written out the way that language says them. The voice model, Supertonic 3 by Supertone Inc., knows more languages than these ten, but we only switch a language on after we have listened to it and checked it with speech recognition.
Can I use the audio commercially?
Yes, on two conditions. The voice model is published under the Open RAIL-M licence, which allows commercial use but forbids a short list of harmful uses — deceiving or harassing people, impersonating someone without their consent and the like; the full list is in section 10 of our terms. And when you publish the audio — a video, a podcast, an advertisement — say that the voice was made by a computer, for example in the description. Every MP3 we make already says so in its own file information.
Why does it sound flat on long texts?
The model reads sentence by sentence, without knowing the story, so it cannot build up drama the way a human narrator does. Short paragraphs with clear punctuation sound noticeably better than one long block — commas and full stops are what it uses to breathe.
Can I change the voice afterwards?
Yes — the pitch and speed sliders cover most of it. Speed is the voice's own speaking rate, so a slower voice simply talks slower instead of turning deeper. For more, download the MP3 and run it through our voice changer, or drop it straight into the audio studio.
How do abbreviations and unusual names get pronounced?
Abbreviations without vowels — PDF, DXF, HTML, GDPR — are spelled out letter by letter in the language you chose, and combinations like MP3 or 4K come out right as well. Numbers, dates, amounts, percentages and clock times are written out the way you would say them. If something still sounds wrong, write it the way it should sound: the voice reads exactly what you type, so an unusual name or brand can simply be spelled phonetically until it sounds right. In Arabic you can also add the vowel marks (tashkeel) where a word needs a particular sound, such as كيفَ or مرحباً; the voice follows them.
