How to Turn a PDF Into a Podcast for Free (2026 Guide)

Updated June 2026.

You have a 40-page lecture PDF and a 35-minute commute. Turning that PDF into a podcast — two AI hosts discussing the material, or a clean narrated walkthrough — is one of the most practical study workflows that has emerged in the last two years. It is also genuinely free to do in 2026, with a few honest caveats about daily limits.

This guide covers the three routes that actually work: Scholarly (our tool — we'll explain exactly what it does and where it fits), Google's NotebookLM Audio Overviews (the most famous option), and a do-it-yourself text-to-speech route using ElevenLabs or OpenAI's TTS API for people who want full control. Then we cover the part most guides skip: what makes a PDF a good source, how long the resulting audio will be, and when audio studying genuinely helps versus when it quietly wastes your time.

Quick answer: what is the fastest free way?

Upload the PDF to a tool that generates a discussion-style episode from it. In Scholarly, that is: create a free account, upload the PDF, click Podcast, and download or stream the episode a few minutes later — then generate flashcards and a quiz from the same PDF to test what you heard. In NotebookLM, it is: create a notebook, add the PDF as a source, and click Generate under Audio Overview. Both have free tiers; both cap how much you can generate per day on free plans.

If you only need the PDF read aloud verbatim — no discussion, no summarizing — a text-to-speech reader is simpler and covered in Method 3.

What makes a good source PDF?

AI podcast generators are only as good as the document you feed them. Before you upload anything, check four things:

• Real text, not scanned images. If you can select and copy text in the PDF, you're fine. If it's a photographed textbook page, run it through OCR first or expect garbled output. (Scholarly and NotebookLM both handle most scanned PDFs with built-in OCR, but clean text always produces a better episode.)
• One coherent topic per episode. A single lecture, a single chapter, a single paper. Feeding a 300-page textbook into one episode forces the AI to compress brutally — you get a vague overview instead of a useful explanation. Split big PDFs by chapter and generate one episode each.
• Prose beats tables and formulas. Dense tables, code listings, and multi-line derivations do not survive the trip to audio. If your PDF is 80% equations, audio is the wrong format for it (more on this below).
• Reasonable length: roughly 5–50 pages. That range gives the generator enough material for a substantive 8–20 minute episode without forcing extreme compression.

Method 1: Scholarly (our tool — here's exactly how it works)

Full disclosure: Scholarly is our product, so read this section as the maker explaining it rather than a neutral review. The honest pitch is that Scholarly is a study workspace, not just a podcast button — the same uploaded PDF also powers cited chat answers, flashcards, quizzes, and AI video lectures, so the podcast becomes one step in a study loop instead of a dead-end MP3.

Step by step:

1. Create a free account at scholarly.so. No credit card required.
2. Upload your PDF. Click Create, choose your file (lecture slides exported to PDF work too). Scholarly extracts and indexes the text.
3. Choose Podcast. Pick the podcast option from the create menu, select your uploaded PDF as the source, and add an optional instruction like "focus on the mechanisms, skip the historical background" if you want the episode steered.
4. Generate and listen. Episodes are typically ready in a few minutes. Stream in the browser or listen on your phone.
5. Close the loop. This is the part we'd argue matters most: from the same PDF, generate a flashcard deck or quiz and test yourself after listening. Listening alone is recognition; answering questions is retrieval — and retrieval is what moves material into long-term memory.

The free plan lets you try this without paying; paid plans (roughly $12–17/month) raise daily creation limits and file-size caps. If you mainly want podcasts plus practice material from one upload, this is the workflow Scholarly is built around. Try it at the AI podcast generator page.

Method 2: NotebookLM Audio Overviews (Google's famous one)

NotebookLM's Audio Overview is the feature that made "turn my PDF into a podcast" mainstream in late 2024, and it deserves its reputation: two AI hosts having a genuinely natural-sounding conversation about your document, free with a Google account.

Step by step:

1. Go to notebooklm.google.com and sign in with a Google account.
2. Create a new notebook and add your PDF as a source (it also accepts Google Docs, Slides, websites, and YouTube links).
3. In the Studio panel, find Audio Overview and click Generate. You can click Customize first to tell the hosts what to focus on.
4. Wait a few minutes, then listen in the browser or download the audio file.

Honest trade-offs: the free tier limits how many Audio Overviews you can generate per day (the paid NotebookLM Plus tier raises limits), the episode length is largely decided for you, and the output is audio plus chat — NotebookLM won't build you a spaced-repetition flashcard deck or a graded practice quiz from the same source. As a pure listening experience, it's excellent. As a complete study loop, you'll be exporting and switching tools.

Method 3: the DIY route — ElevenLabs or OpenAI TTS

If you want total control over voice, pacing, and script — or you're a tinkerer — you can build the pipeline yourself. This route produces narration (one voice reading a script), not a two-host dialogue, unless you script the dialogue yourself.

1. Extract the text. Copy it out of the PDF, or use a converter for big files.
2. Turn the text into a script. Pasting raw textbook prose into TTS produces stilted audio. Ask an AI chatbot to "rewrite this chapter as a clear 10-minute spoken explanation for a student" first.
3. Generate the audio.
   • ElevenLabs: the free tier includes a monthly credit allowance (roughly ten minutes of audio) with very natural voices in dozens of languages. Their ElevenReader app can also turn documents into GenFM podcast-style episodes directly.
   • OpenAI TTS API: around $15 per million characters for the standard model — a full lecture's worth of audio costs well under a dollar — with around a dozen preset voices. Requires writing a small script and an API key, so this is the developer option.
4. Move the MP3 to your phone and listen.

The DIY route wins on control and per-minute cost at scale; it loses badly on convenience. For one-off studying, Methods 1 and 2 are ten times faster.

How long will the podcast be?

Set expectations correctly, because this surprises people:

• A 10–20 page lecture PDF typically becomes an 8–15 minute episode.
• A 40–50 page chapter becomes roughly 15–25 minutes — the AI compresses, it does not read everything.
• A 200-page textbook does not become a 3-hour faithful audiobook in any of these tools. Split it into chapters and make a series.

Compression is a feature, not a bug: a good study episode is a guided tour of the important ideas, not a verbatim reading. But it means you should never treat an AI podcast as complete coverage of the source. Details, edge cases, and exact figures stay in the PDF.

When does audio studying actually work — and when doesn't it?

This is where we'll be more honest than the average tool blog. The research framing that gets cited for study podcasts is usually dual coding theory — Allan Paivio's finding that information encoded both verbally and visually is remembered better than information encoded one way. Strictly speaking, listening to a podcast instead of reading is not dual coding; it's swapping one verbal channel for another. The honest version of the claim is this: audio works best as a second, spaced exposure to material you have already met visually — lecture, slides, reading — because revisiting the same ideas through a different presentation, at a different time, strengthens retrieval routes. That's spaced review doing the work, with audio as the convenient delivery mechanism.

Audio studying works well when:

• You're reviewing, not learning from scratch — the episode reactivates things you've seen before.
• The material is conceptual: mechanisms, arguments, frameworks, case studies, historical narratives.
• You'd otherwise be doing nothing — commutes, walks, gym sessions, dishes. Recovered dead time is the genuine superpower here.
• You follow it with retrieval practice — flashcards or a quiz on the same material — because listening by itself is passive and famously easy to tune out.

Audio studying works poorly when:

• The material is equation- or diagram-heavy (math derivations, organic chemistry structures, circuit analysis). Audio is transient; you can't stare at line three of a derivation.
• It's your first contact with dense, unfamiliar material. Working memory overloads fast when you can't slow down and re-read.
• You treat listening as the whole study session. Familiarity from passive listening feels like knowledge but doesn't survive an exam. Listen, then test.

FAQ

Can I really do this for free?

Yes. Scholarly's free plan, NotebookLM's free tier, and ElevenLabs' free monthly credits all produce real episodes at zero cost. Every free tier has daily or monthly generation caps, so "free" realistically means a handful of episodes per day — fine for studying, limiting if you're batch-converting a whole textbook in one sitting.

What languages are supported?

NotebookLM generates Audio Overviews in dozens of languages. ElevenLabs supports several dozen languages across its voice models. Scholarly can generate podcasts from sources in major languages as well. For all tools, output quality is strongest in English; if your PDF is in another language, generate a short test episode before committing to a full series.

Can I choose the voices?

It depends on the route. The DIY route gives you the most choice — ElevenLabs has a large voice library and OpenAI offers a set of preset voices. NotebookLM uses its signature two-host pair with limited control. Scholarly provides tuned host voices optimized for clear explanation rather than a large pick-list.

Will the podcast cover everything in my PDF?

No — and no tool honestly claims it will. Expect a compressed guided tour of the main ideas. Use the episode for review and the original PDF (or cited chat answers over it) for details.

PDF is scanned — will it work?

Usually. Scholarly and NotebookLM run OCR on scanned documents. Very low-quality photo scans of handwriting are the main failure case; if the text extraction is garbage, the audio will be too.

What's the best PDF-to-podcast workflow for exam revision?

Split the syllabus into chapter-sized PDFs, generate one episode per chapter, listen during dead time across the week, and pair each listen with a 10-question quiz on the same chapter. That pairing — spaced audio re-exposure plus retrieval practice — is the configuration the learning science actually supports, and it's the loop Scholarly's podcast generator is designed around.