EPUB to Audiobook
Turn a DRM-free ebook into an audiobook narrated by a natural-sounding AI voice. You get one MP3 per chapter, or a single M4B with chapters. The voice model runs on your own device, so the book is never uploaded.
✓ Files never leave your device ✓ Works offline ✓ No signup, no watermark
Drop an EPUB, PDF, or TXT here
or click to browse
English text works best. DRM-protected Kindle and Apple Books files can't be read.
Hear the voices first. Tap one to play a sample, no download needed:
How it works
- Drop the book. Chapters appear with word counts. Front matter like copyright pages is unticked automatically.
- Pick a voice and press Preview. The first time, your browser downloads the voice model (92 MB) and keeps it for next time. The preview also measures how fast your device is, so the time estimate is real.
- Press Create audiobook. Chapters are narrated one after another. Each finished chapter is saved in your browser right away, and you can play it immediately.
- Download one MP3 per chapter as a zip, or a single M4B audiobook with chapter marks for Apple Books, BookPlayer, or any audiobook app.
The voice
Narration uses Kokoro, an open-weight text-to-speech model released under the Apache 2.0 license. With only 82 million parameters, it's small enough to run in a browser, yet in listening tests it ranks alongside much larger commercial voices. It handles long passages with natural pacing and intonation, which older browser voices never managed. The list above includes only the voices Kokoro's authors grade highest; Heart is the most natural. Kokoro is trained mainly on English. Other languages will either sound wrong or be read with an English accent.
How long it takes
Everything runs on your computer's processor, so speed depends on your hardware, and it varies a lot between machines. That's why the preview measures your device before you commit, and the status line shows a live time estimate while it works.
Plan on it being a background job. A few chapters are a coffee break. A whole novel is something to start before bed. It's slow because the narration is actually being computed on your machine instead of a rented server farm, and that's also why it's free and private.
You don't have to do it in one sitting. Press Stop or close the tab, and every finished chapter is kept in this browser. Open the same file later and press Continue. While it's generating, the page keeps your screen from sleeping.
Why on-device matters here
Online "ebook to audiobook" services upload your whole book to their servers, and most charge by the character. A 90,000-word novel is about half a million characters, which costs from a few dollars on basic cloud voices to well over $100 on premium ones. Here, nothing leaves your computer: the text is split into sentences in this tab, the voice model turns them into audio on your own processor, and the MP3s are written straight to your downloads. That makes it suitable for unpublished manuscripts, work documents, course readers, and anything else you'd rather not hand to a third party. You can check: once the model has downloaded, turn off Wi-Fi and generation carries on.
What to use it for (and not)
- Great for: public-domain classics from Project Gutenberg and Standard Ebooks, your own manuscripts (hearing a draft read back is the best proofreading there is), long reports and papers to listen to on a commute, and accessibility for readers with dyslexia or low vision.
- Not for: DRM-protected purchases, which can't be opened, or redistributing narrations of copyrighted books you don't own the rights to. Personal listening is between you and the book's license. Publishing is a different matter.
Common problems
- "No text found" on a PDF. It's a scan. Run it through Searchable PDF first, then bring the result here.
- Headers, page numbers, or footnote markers are read out. These are ordinary text in a PDF. For cleaner narration, convert with PDF to EPUB first, or untick chapters that are mostly notes.
- The download stalls on the first run. The model comes from Hugging Face's CDN. Corporate networks and some ad blockers block it. Try another network, or allow
huggingface.co. - It's very slow. Close other heavy tabs and apps, and plug your laptop in. Battery-saver modes throttle the processor hard.
- A name or unusual word is mispronounced. The model guesses pronunciation from spelling. It gets common words right and rare proper nouns sometimes wrong. There's no pronunciation editor yet.
Frequently asked questions
Is it really free? What's the catch?
Yes. The heavy lifting happens on your device, so it costs us nothing to run. The site is supported by ads. No signup, no watermark, no character limit.
Is my book uploaded anywhere?
No. The only downloads are the voice model and engine code, both public files that are the same for everyone. Your book's text is never sent anywhere; watch the Network tab in DevTools while it runs to confirm.
Can I choose different voices for different characters?
Not yet. One voice narrates the whole book. You can use a different voice per chapter by changing the voice between runs, since finished chapters are stored per voice.
What format are the files?
MP3, mono, 64 kbps at 24 kHz: about 29 MB per hour, and plenty for a single voice. The M4B option re-packs those chapters into one AAC audiobook file with chapter marks, cover-less, titled with the book's metadata.
Does it work on a phone?
It works, but slowly, and phones pause background tabs. For anything longer than a chapter or two, use a laptop or desktop and send the finished files to your phone.
I just want to listen right now, without waiting.
Use Read Aloud. It starts instantly with your browser's built-in voices, though it can't save a file.