Blog ›

How to have a PDF read aloud (text to speech) without the page numbers, footers and gibberish

30 September 2026 · 8 min read

You want to listen to a report on the train, get through a 60-page paper while cooking, or you simply read better by ear. Every device you own can already read a PDF aloud. The catch is that PDFs were designed to be printed, not spoken, and a reader that starts with "Page 3 of 12, Confidential, Annual Report 2025" every thirty seconds is unbearable. Here is what works on each device, how to make an audio file you can keep, and the five reasons PDFs sound wrong, with the fix for each.

The readers you already have

iPhone and iPad

Turn on Settings → Accessibility → Spoken Content → Speak Screen. Open the PDF in Files, Books or Safari and swipe down from the top of the screen with two fingers. A small controller appears with play, speed and skip. Under Voices, download one of the enhanced or premium voices for your language once; the default compact voice is the reason people think text to speech sounds robotic. The same menu has Speak Selection for reading just a highlighted passage.

Mac

Open the PDF in Preview, select the text (or ⌘A for everything), then Edit → Speech → Start Speaking. Or enable System Settings → Accessibility → Spoken Content → Speak selection and press Option-Esc anywhere. Download a Siri or premium voice in the same place. Preview reads in the document's text order, which is the source of most of the problems below.

Windows

The best free PDF reader-aloud on Windows is the browser: open the PDF in Microsoft Edge, right-click, Read aloud (or Ctrl+Shift+U). Edge's natural online voices are excellent, it highlights the sentence being spoken, and it lets you click any paragraph to jump there. Adobe Acrobat also has View → Read Out Loud, with older voices.

Android

Settings → Accessibility → Select to Speak reads whatever is on screen after you tap the button and drag over the text. Google's Speech Services voices are good; check for an updated voice pack under Text-to-speech output. Google Play Books and Chrome's "Listen to this page" also work with PDFs you open in them.

Dedicated apps

Speechify, NaturalReader, Voice Dream Reader (iOS/Mac), @Voice (Android) and ElevenLabs Reader offer better voices, sentence highlighting, per-document speed and, importantly, some layout clean-up such as skipping headers. They are worth paying for if you listen daily; for occasional use the built-in readers are enough. Note that most of them upload the document to generate the voice.

Making an MP3 you can keep

Listening live is fine for one document; for a commute you want a file. Three routes:

• Mac, no install: the say command writes audio. Export the PDF's text first (Preview: File → Export as Text, or pdftotext -layout in.pdf out.txt), clean it (see below), then:

  say -v "Samantha (Enhanced)" -f out.txt -o report.aiff
  ffmpeg -i report.aiff -b:a 96k report.mp3

• Any platform, best free voices: the open-source edge-tts package uses the same neural voices as Edge's Read Aloud: pip install edge-tts, then edge-tts --file out.txt --voice en-GB-SoniaNeural --write-media report.mp3. This does send the text to Microsoft's service.
• Fully offline: Piper (open source, runs locally, good voices in many languages) if the document is confidential and must not leave your machine.

The step everyone skips is the middle one: clean the text. Whatever you feed a voice, it will read. Which brings us to why PDFs sound wrong.

Why PDFs read aloud badly, and how to fix each cause

1. Headers, footers and page numbers are read on every page

The single biggest annoyance. "Confidential. Quarterly Report. Page 7." twelve times. Readers can't tell a running header from body text because in the PDF it's all just text at coordinates. Fix: in the exported text, delete every line that repeats on every page. A find-and-replace for the header text and a regular expression for ^Page \d+( of \d+)?$ handles most documents. Apps like Voice Dream let you set a "skip lines shorter than N characters" rule, which catches the same thing.

2. Multi-column layouts are read across, not down

Academic papers, magazines and newsletters put two or three columns on a page. Many readers take the text line by line across the page, so you hear half a sentence from column one, then half a sentence from column two. Fix: use a tool that respects columns when extracting. pdftotext without -layout usually gets columns right; so does exporting from Acrobat with "reflow" on, or converting the PDF to EPUB (Calibre's Heuristic processing) and listening to that instead.

3. Words hyphen-ated at line ends become two words

Justified text breaks "information" into "infor-" and "mation" on separate lines; the voice reads "infor. mation". Fix: a regular expression that joins them: replace -\n with nothing (or with nothing only when the following line starts lowercase, to keep real hyphens like "well-known"). Calibre's EPUB conversion does this automatically.

4. Tables, figures and footnotes turn into number soup

A financial table read as a stream of digits is meaningless; a footnote marker sends the voice into the footnote and back. Fix: decide which you want. For a listen-through, delete tables and footnotes from the text and keep the sentence that introduces them ("Revenue by region is shown in Table 2"). If you need the table, rewrite it as sentences ("Germany: 4.2 million, up 6%") — a summarizing tool can do this for you.

5. Scanned PDFs have nothing to read

If you can't select text with your mouse, the PDF is a picture and the reader stays silent or reads garbage. Run OCR first: Acrobat's Recognize Text, the free ocrmypdf, or on iPhone simply open the scan in the Files app, where Live Text lets Speak Screen read it. OCR errors ("rn" read as "m", "1" as "l") are far more noticeable by ear than by eye, so listen to the first page before committing an hour to it.

A quick pre-listen routine

1. Export the text (Preview, Acrobat or pdftotext).
2. Delete headers, footers and page numbers (repeating lines).
3. Join hyphenated line breaks.
4. Remove or rewrite tables; drop footnotes unless you need them.
5. Listen to the first two minutes at your normal speed. If it's wrong, it's wrong everywhere.

Ten minutes of preparation turns an unlistenable hour into a comfortable one. If that sounds like a lot for one document, that is exactly the job we'd like to hand to an agent.