To turn a PDF into speech, open PDF to Audio, upload the file, pick a voice, and download an MP3 or WAV. File Converts extracts the text and synthesizes it with a neural voice that runs on your device — so your document is never uploaded.
Why text-to-speech is useful
Listening lets you get through reports, articles, and study material while commuting, walking, or resting your eyes — and it's a genuine accessibility aid. The catch with most TTS services is that they upload your text and meter every character. An on-device voice removes both problems.
On-device, natural voices
File Converts runs an 82-million-parameter neural voice model (Kokoro) directly in your browser via WebAssembly. The model downloads once on first use, then synthesizes locally — your text never leaves the device, and there's no per-character cost or limit.
How to convert
- From a document: PDF to Audio or Word to Audio — the text is extracted, then spoken.
- From typed or pasted text: Text to Speech.
- Pick a voice and export MP3 or WAV; the audio is generated on your device and downloads instantly.
Tip: get clean text first
For a scanned PDF (an image of text), run Image to Text (OCR) or PDF to Text first so the words are recognized, then feed that into text-to-speech for the best result.
What to expect the first time you run it
Because the voice model runs on your device rather than on a server, it has to get to your device first. The initial visit downloads the model once — a wait of anywhere from a few seconds to a couple of minutes depending on your connection — and the browser then caches it. Every run after that starts immediately, and works with no network connection at all.
Synthesis speed depends on your hardware, not on a queue. A page or two is quick on any modern laptop or phone; a long report takes proportionally longer, and the work happens in a background worker so the page stays responsive while it runs. There is no character cap and no per-month allowance, because nothing is being metered on the other end.
What the voice does with document structure
Text-to-speech reads words, not layout. Headings are spoken as ordinary sentences, so a document flows as continuous prose rather than announcing its sections. Page numbers, running headers and footers are part of the extracted text too, and they will be read out wherever they fall — usually mid-sentence, at every page break.
Tables and multi-column pages are where this shows most, because the text is extracted in reading order and a two-column layout can interleave into something that makes no sense aloud. Images, charts and equations are skipped entirely; they carry no text to speak. For a long or heavily formatted document it is worth running PDF to Text first, tidying the output, and feeding the clean version to Text to Speech.