Read your PDFs aloud, the way a human would

A PDF doesn't store paragraphs. It stores glyphs at coordinates, and the line breaks you see are a consequence of the page width, not of the sentence. When a naive reader pulls text out of a PDF it gets one line at a time, and it speaks them one line at a time — which is why so many apps read a research paper with a small hesitation every nine words, and pronounce a word split across a line break as two separate words.

Sandbook repairs this before a word is spoken. Visual line breaks are rejoined into continuous prose, and words hyphenated across a line end are stitched back together. "Represen-" plus "tation" becomes representation, not "represen" then "tation". The result is a PDF that reads like a page of a book instead of a list of lines.

It is the specific reason PDF to speech works better here than in readers that treat a PDF like a text file — an EPUB already tells you where its paragraphs end, and a PDF never does.

One expectation to set before you install: Sandbook is a reader, not an audiobook exporter. It reads a PDF aloud while you have it open; it does not convert a document into an audio file you can keep. M4A export exists, but only in the Reader tab for text you typed or pasted there. If what you want is a whole book turned into a file, this app does not do it.

1. Open the file in Sandbook. Use the Files picker inside the app, or choose Open with Sandbook on a PDF in Files, Mail or Safari. The document is copied into the app's storage, so it stays readable after the original moves.
2. Find it on the shelf. PDFs show a thumbnail of their first page, the title, a format badge and a progress bar.
3. Press play. Text is extracted a page at a time, so a 600-page scan of a textbook doesn't try to load itself into memory at once. Speech starts after the first sentence and keeps generating ahead of you.
4. Move around. Tap any paragraph to start reading from there, skip forward or back, or open the page list and jump to page 214.

A PDF's layout is often the point — figures, equations, tables, a diagram you need to look at while the caption is read. So Sandbook shows the real page through PDFKit, with pinch-zoom and page-curl paging, rather than flattening your document into a wall of extracted text.

While it reads, the current paragraph is highlighted and the view scrolls to follow (word-by-word marking is for the text reader, not this page view), so you can track the argument visually while listening. When a chapter ends the reader carries on to the next page by itself, and your place in the document is saved per file, so you can put a 40-page report down and come back to page 23 tomorrow.

If your PDF is a photograph of paper, a scanned book or a fax, there is no text layer in the file, and Sandbook does not run text recognition on PDF files. It will open the document and find nothing to say.

Paper itself is a different story. Since version 1.0.4, Sandbook can scan printed pages with the camera, up to 20 at a time, or read a photo or screenshot you pick. The text is recognized on your iPhone and each page becomes a chapter you can listen to. So for a handout or a book chapter on paper, scan the paper; for an image-only PDF, export its pages as images and use From Photo.

Apps that run text recognition on PDF files directly, such as Speechify and NaturalReader, are the better tool if most of your documents are image-only PDFs.

Reading a paper aloud is a different job from reading a novel. Speed runs 0.5× to 2.5× in tenth steps, applied by scaling predicted phoneme durations inside the model rather than by resampling the audio, so pitch is preserved — 2× is a person reading quickly, not a chipmunk, and 0.7× on a difficult passage stays natural instead of sounding drugged.

For technical PDFs, the steadier voices tend to work best. Every one of the 41 voices can be auditioned live in the app with the real model before you commit forty minutes to it, and starred so it floats to the top next time. Before installing, you can try a paragraph from your paper in the browser.