Skip to content
audiobook.tools
AI

PDF to Audiobook

Turn a PDF into an audiobook read by a natural AI voice, with the page numbers, running headers and footnotes taken out and chapters from its bookmarks. Free, and the PDF never leaves your device.

  • Free
  • No upload
  • No sign-up

The AI voice runs in your browser: it downloads once, then works offline · nothing is uploaded

Where it runs

Works in a current Chrome, Edge, Firefox or Safari, on Windows, Mac, Linux and Chromebook computers, Android phones, iPhones and iPads. Nothing to install, no account.

The first narration downloads the voice model once: about 93 MB on a processor, or 327 MB with a graphics card (WebGPU, in current Chrome and Edge), which is several times faster.

The first file loads the audio engine (ffmpeg, about 31 MB) once, and your browser keeps it. Very long books are easiest on a computer, which has more memory than a phone.

Some PDFs need standard character and font tables, fetched from a public CDN; your PDF stays on the device.

How to use it

4 simple steps.

  1. 01

    Drop your PDF

    Any PDF with selectable text: a manuscript, a report, a public-domain book. It’s read page by page on your device; a password-protected PDF asks for its password here.

  2. 02

    Check what will be read

    Chapters come from the PDF’s bookmarks or its headings. Running headers, page numbers and footnotes are taken out, and words hyphenated across lines joined. Pick the pages and preview the cleaned text.

  3. 03

    Pick a voice

    Ten natural voices, American and British. Tap an orb to hear it, set the speed, and hear a paragraph of your own PDF.

  4. 04

    Narrate and download

    Chapters are voiced one after another in your browser and saved as they finish. Download an M4B with chapters and a cover, or one MP3 per chapter.

What happens to your PDF

A PDF doesn’t store paragraphs. It stores runs of text and where each one sits on the page, which is why copying text out of a PDF so often brings page numbers, broken lines and half-words with it. Before a voice can read it, the page has to be rebuilt the way a reader sees it:

StepWhat it does
Lines and columnsRuns on the same baseline become a line; a two-column page is read down each column in turn. Ligatures (fi, fl) become plain letters.
Running headers and footersA line near the top or bottom that repeats across the pages, with only its page number changing, is left out.
Page numbers“12”, “xii”, “Page 12 of 300” and “— 12 —” at the edge of a page are left out.
FootnotesSmall type below the last line of body text, and the raised markers pointing to it, are left out unless you ask for them.
ParagraphsA new paragraph starts at an indent, a wider gap, a change of type size, or after a short line that ends a sentence. A paragraph carries on over a page break.
Hyphenation“narra-” at a line’s end and “tion” on the next become “narration”; a real compound like “Anglo-Saxon” keeps its hyphen.
ChaptersFrom the bookmarks, or from headings in bigger type and “Chapter” lines. A drop cap (a big first letter) joins its word.
SpeechEach paragraph is tidied for the voice (abbreviations, numbers, symbols, web addresses), then read by Kokoro-82M on your graphics card or processor.

Which PDFs work best

  • Best: PDFs exported from Word, Google Docs, Scrivener, InDesign or Vellum, and ebooks saved as PDF. They have clean text and usually bookmarks.
  • Good: reports, essays and public-domain books from archives that include a text layer. Check the preview: old OCR can carry spelling errors that the voice will read as written.
  • Hard: magazines, brochures and slide decks (text boxes in no fixed order), and pages heavy with tables, maths or code.
  • Won’t work: scans without a text layer, and DRM-locked PDFs.

Have the manuscript as a Word file or EPUB too? Those keep their structure better than any PDF: use text to audiobook or EPUB to audiobook. To estimate the finished length first, the manuscript analyzer counts words and chapters, and the pronunciation guide maker collects names the voice should say right.

Where it runs

WhereDownload (once)Speed
Graphics card (WebGPU)about 330 MBusually faster than real time
Processor (WebAssembly)about 93 MBabout 2–3 × the audio’s length

The voice model is cached by your browser after the first download, so later books start straight away and work offline.

FAQ

Questions, answered.

Is my PDF uploaded anywhere?

No. pdf.js, the open-source PDF reader inside Firefox, reads the file in your browser, and the AI voice (Kokoro) runs on your device. Two things download, never your file: the voice model, from Hugging Face, once, and for PDFs whose fonts need them, standard character tables from a public CDN.

Why does it say my PDF is scanned?

A scanned PDF holds pictures of pages, not text, so there is nothing for a voice to read. Adobe Acrobat, most scanner apps and many free tools can add a text layer with OCR (optical character recognition). Run that, then drop the new PDF here. A quick test: if you can select and copy words in your PDF viewer, it will work.

Will it read the page numbers and headers?

No. Lines that repeat at the top or bottom of the pages (a book’s title, a chapter’s name, “Page 12 of 300”) are recognised and left out, as are footnotes in small type at the foot of a page and the little numbers that point to them. The panel says how many of each were removed. Tick “Read footnotes aloud” to keep footnotes.

Where do the chapters come from?

From the PDF’s bookmarks (its outline) when it has them: that’s how most ebooks and many exported documents mark chapters. Otherwise from headings, found by their bigger type or by lines like “Chapter 3”. You can switch between the two, or read the whole range as one piece, and rename or untick any chapter. The pages before the first chapter (title page, copyright, contents) start unticked.

Can it handle two columns, textbooks and papers?

Two-column pages are read down the left column, then the right. Tables, equations, code and figure captions are read as plain text, in the order they sit on the page, which rarely sounds good: leave those pages out with the page range, or untick their chapters. For layouts that need a lot of fixing, the script cleaner lets you edit the text before narrating it.

What about a password-protected or DRM-protected PDF?

If a PDF opens with a password, type it in when asked: it’s used on your device to open the file and goes nowhere else. PDFs locked to a store’s reader app (DRM) can’t be opened, and shouldn’t be: you need the rights to turn a text into audio.

How long does it take?

Reading the PDF takes seconds. Narration depends on your device: with WebGPU (current Chrome and Edge on most computers) it usually runs faster than real time; on a processor alone it takes about two to three times the length of the audio. A 300-page novel (about 90,000 words) makes roughly 9 to 10 hours of audio. You can pause, close the tab and resume later: finished parts are saved in your browser.

Can I sell an audiobook made from a PDF with an AI voice?

The voices are free to use commercially under the Apache 2.0 licence, and you need the rights to the text. Stores set their own rules for AI narration: our AI audiobooks guide lists each one’s, with sources.

Keep going

Related tools.