"Audiobook from PDF" used to mean firing up a clunky desktop converter, exporting an MP3, and side-loading it onto your phone. In 2026 the entire pipeline lives in a single iPhone app. This guide covers the modern workflow, when to use which app, and a few things almost nobody mentions.
The fast path (60 seconds)
- Install a TTS reader. Murmura is free, NaturalReader and Voice Dream are paid alternatives.
- Import the PDF — from Files, from a URL, from an email attachment, or from a built-in Project Gutenberg / Internet Archive catalog.
- Tap Listen. The narration streams within a second.
That's it. The audio is generated on demand and cached locally after the first play, so re-listening works offline.
The slow path (export as MP3)
You might want an actual audio file — to email to someone, drop into a podcast feed, or load onto a non-Apple device. Few iOS apps export to MP3 directly because Apple's review rules are tetchy about it. Options that do:
- NaturalReader Plus (paid) exports to MP3 on iOS.
- Speechify Premium exports to MP3 on the web client.
- Microsoft Azure Speech directly, if you're technical enough to call an API. The voices in Murmura are the same family.
Handling difficult PDFs
Scanned PDFs (no embedded text)
Most TTS apps assume the PDF has real text inside, not a photo of text. If your PDF was scanned, you need to OCR it first. The fastest free option: open the PDF in Apple Notes — iOS will OCR it automatically — then export back to PDF. Adobe Acrobat and Microsoft Lens also work.
Multi-column academic papers
Two-column papers used to be a nightmare. Modern parsers (Murmura, Speechify, NaturalReader) handle column flow correctly by detecting reading order. Older apps may read across columns and produce nonsense.
Math-heavy PDFs
Math is still hard. LaTeX-rendered formulas usually get skipped or read out as gibberish. The best you can do today: import the PDF, skip math sections manually, and look at the formulas on the screen while the surrounding prose narrates.
Books with footnotes
Footnotes embedded inline get read in the wrong order. Look for an app that detects footnotes and queues them at section boundaries. Murmura does this for footnote markers it recognises; complex academic typography is still a work in progress.
Voice selection matters
The voice changes how much you absorb. A few rules of thumb:
- Technical content: a calm, clearly-articulated voice. Murmura's Andrew or Jenny. Speechify's "Snoop" or "Mark Zuckerberg" voices are jokes — don't use them for real reading.
- Fiction: a voice with expressive range. Murmura's Sonia (British) or Emily (Irish). ElevenLabs voices excel here.
- Reference / quick scanning: a faster voice at 1.6–2×. Listen for clarity loss at that speed before committing.
Speed matters more than voice
Almost every reader settles into a personal speed somewhere between 1.4× and 1.8× after a week of use. That's roughly twice the speed of normal speech — slow enough to follow, fast enough to cover real ground. If you cap at 1× you'll never finish anything.
Privacy considerations
Most TTS apps send your PDF text to a cloud service for neural generation. That's fine for public-domain books, less fine for confidential work documents or unpublished research. Check the privacy policy before importing anything sensitive. Murmura processes text on a server but deletes it immediately and retains only an irreversible hash for caching. Voice Dream uses iOS system voices by default — those run entirely on device.
Cost-per-book reality check
A 100,000-word book at $2.99/month Pro is roughly $0.10 of synthesis cost. At Speechify's $11.58/month effective rate, it's closer to $0.40. The actual margin is reasonable for an indie developer at the Murmura price point and generous at Speechify's. You are not, in 2026, paying for compute. You are paying for the product layer.