I spent three hours last Tuesday trying to listen to a 40-page research paper while making dinner. Three hours. Not because the paper was hard to convert — because every tool I tried was broken in a different way.
Adobe Acrobat's "Read Aloud" went first. View menu, Read Aloud, Activate Read Aloud. It started talking in that voice. You know the one. The GPS navigator voice that pronounces "albeit" as "all-bite" and reads "Figure 3.2: Distribution of p-values across trials" with the same flat affect as the actual findings. It also read every page number, every running header, and every footnote marker. "Page seventeen. Bayesian Methods in Clinical Research. Superscript forty-two." No.
Then NaturalReader. Uploaded the PDF. "Free users can convert up to 5 minutes of audio." The paper would take about 50 minutes. So that's $60 per year. For PDFs. That I already own. I closed the tab.
Speechify. Same thing. Upload, start listening, and — "Upgrade to Premium for unlimited listening. $139/year." I don't want unlimited listening. I want to listen to one paper about Bayesian priors while I chop onions. $139.
So I tried the old reliable: copy-paste into Google's TTS demo. Except the PDF was two columns. I selected all text, pasted it, and got the left column of page 1 followed by the right column of page 1, then left column of page 2... actually no. I got a garbled mess where sentence fragments from column A were interleaved with sentence fragments from column B. "The model was trained on participants who 3.2 Methods We recruited." I tried selecting column by column. That worked for about two pages before I realized I'd be doing this for 20 pages and it was now faster to just read the paper with my eyes.
The problem is that PDFs aren't text files. This is the thing nobody tells you. A PDF is a set of instructions for placing ink on a page. "Put this glyph at coordinates (72, 684). Put this glyph at coordinates (78, 684)." There's no paragraph structure. There's no reading order. There's just ink coordinates. When a tool says "extract text from PDF," what it's actually doing is guessing where the paragraphs are based on the spatial positions of characters. Two-column layouts break that guess. Headers and footers are indistinguishable from body text. Page numbers look like any other number.
CastReader handles this differently. I uploaded the same paper — the same two-column, footnote-heavy, figure-laden paper — and it came back clean. Left column before right column. No page numbers in the narration. No "Figure 3.2 colon" interruptions. The footnotes were gone. The running header was gone. What came through was the actual content, in order, read by an AI voice that sounds like a person who's read a lot of academic papers and knows where to pause.
The voice is Kokoro, an open-source neural TTS model. Not the robotic Speak & Spell voice that Adobe uses. Not great for audiobook narration of literary fiction — it's not Morgan Freeman. But for a research paper, a textbook chapter, a contract, a government report? It's more than good enough. You stop noticing the voice after about two minutes and start actually absorbing the content. I've listened to maybe 30 papers through it now. The comprehension is genuinely better than screen reading when I'm doing something else with my hands.
The practical stuff: you upload at castreader.com/pdf-to-audiobook. Drop the file, wait maybe 20 seconds, and it opens a reader with each chapter as a separate track. You can play in the browser or download MP3s. There's no page limit. I tested a 600-page intro textbook on partial differential equations — it took about 90 seconds to process and produced 47 chapter tracks. My colleague uploaded her entire Federal Rules of Civil Procedure PDF. Worked fine.
It's free. Actually free. Not "free for 500 characters" or "free for your first document" or "free with a watermark." Free. No account. I don't even think they know who I am — there's no login.
For PDFs you read regularly online — arXiv papers, government documents, legal databases — there's another option that's honestly even better. The CastReader Chrome extension reads PDFs directly in your browser tab. You don't download the file, you don't upload it anywhere. You open the PDF in Chrome, click the extension icon, and it starts reading with paragraph highlighting that follows along on the page. I use this for arXiv now. Open paper, click icon, listen while I walk to get coffee. One click.
The extension also works on regular webpages, Kindle Cloud Reader (it's the only one that can — Amazon encrypts the text), Google Docs, basically everything. But for PDF-to-audiobook specifically, the upload tool at /pdf-to-audiobook is the thing.
Scanned PDFs are trickier. If your PDF is just images of pages — an old textbook scan, a photographed document — CastReader runs OCR on it. The results depend on scan quality. A clean 300-DPI scan of a modern textbook? Near perfect. A grainy photocopy of a 1970s journal article? Maybe 80% accurate, with some garbled words. It's honest about this — there's no magic that makes bad scans perfect. But for most scanned documents people actually encounter, it works well enough.
Languages are handled automatically. Upload a PDF in Japanese, French, Arabic, Korean — it detects the language and picks a native voice. A friend of mine is learning Portuguese and uploads Brazilian news PDFs every morning. She reads along while listening. She says her pronunciation improved more in two months of this than in a year of Duolingo. I can't verify that claim but she's very enthusiastic about it.
Here's what I tell people when they ask about PDF-to-audiobook conversion in 2026. If you need it once for a quick document, use the upload tool at castreader.com/pdf-to-audiobook. If you read PDFs online regularly, install the Chrome extension — it's faster and you don't have to download anything. If you want to pay $60-$139 per year for essentially the same output, NaturalReader and Speechify are right there. And if you want the authentic 2003 robot voice experience, Adobe Read Aloud is free and built in.
I finished that research paper, by the way. Listened to the whole thing while making pasta. Understood maybe 70% of the Bayesian methodology, which is about what I'd get from reading it on screen. The onions turned out great.
Related Reading
More free audiobook and document TTS tools.
