EPUB to MP3, M4B or MP4
Some EPUBs carry a full recorded narration that most reader apps will not play. This tool pulls that audio out into regular MP3, M4B or MP4 files you can play anywhere — and for text-only books, it reads the text aloud with on-device text-to-speech. Everything runs in your browser.
- DRM-protected books cannot be converted.
- Text-to-speech downloads a voice model once (size shown before download).
Drop an EPUB file here, or click to choose one
EPUB 2 / EPUB 3, with or without recorded audio
Book
Chapters
Choose what to do
The voice model (Supertonic 2) is about 250 MB. It downloads only when you press the button, is cached by your browser, and synthesis then runs entirely on your device.
The read-aloud voice is AI-generated speech. Do not use it to impersonate or deceive others.
Output
Use the audio you make for your own listening. Do not share or upload audio or video made from a book you do not hold the rights to.
How it works
1. Add your EPUB
The file is opened locally: metadata, cover, chapter list, any recorded audio, and the full text are read in your browser.
2. Extract or synthesise
The original audio is reused whenever possible — MP3 files are joined without re-encoding, and AAC audio is remuxed into M4B without quality loss. Text-to-speech (Supertonic 2) generates audio chapter by chapter on your device and can resume where it stopped.
3. Take your file
MP3 with chapter tags, M4B with chapters and cover art for audiobook apps and Apple Books, or an MP4 video that shows the cover while the audio plays.
Good to know
- Video (MP4) uses WebCodecs and works in current Chrome and Edge. Without it, you can still download the M4B and the cover image separately.
- For books over 500 MB, or on a phone, splitting into one file per chapter is recommended. Keep the tab in the foreground on phones — mobile browsers pause background tabs.
- AAC encoding also uses WebCodecs. On browsers without it, AAC originals are still remuxed losslessly, and MP3 output remains available.
FAQ
Why can I not hear the narration in my reader app?
EPUB3 media overlays are optional for reader apps, and many do not implement them. The audio files are inside the book, though — this tool extracts them into ordinary audio files.
Is my book uploaded anywhere?
No. The EPUB is parsed and converted entirely in your browser. Nothing about the file or its text is sent to a server.
Can I convert a book I bought from a store?
If the book has no DRM, yes — for your own listening. DRM-protected books are detected and refused, and this tool never removes or bypasses DRM.
Why does text-to-speech ask to download a model?
The voice (Supertonic 2, about 250 MB) runs locally on your device, so the model is fetched once — only when you press the download button — and cached by your browser. It is never downloaded again unless you clear site data.
Is the read-aloud voice a human recording?
No — it is AI-generated speech, and files made with it are labelled as AI-generated speech in their metadata.
Do chapters survive?
Yes, when “Include chapter markers” is on: M4B carries a chapter list, MP3 carries chapter tags, and you can also split the book into one file per chapter in a ZIP.
Open-source licenses
This tool is built with the components below. Heavy components load only when the matching feature is used.
- Supertonic 2 (Supertone) — OpenRAIL-M (weights) · MIT (code). On-device text-to-speech. The model runs in your browser from the pinned release; usage restrictions are summarised below.
- ONNX Runtime Web 1.17.3 — MIT. Runs the speech model (WebGPU or WASM), loaded only for text-to-speech.
- fflate 0.8.2 — MIT. ZIP reading/writing for EPUB files
- lamejs 1.2.1 — LGPL-3.0. MP3 encoding, loaded dynamically only when MP3 re-encoding is needed. Source: github.com/zhuker/lamejs (a port of LAME, LGPL). No GPL components are used.
- WebCodecs (browser built-in) — platform. AAC/H.264 encoding where the browser provides it
About the speech model (Supertonic 2)
Speech is generated with Supertonic 2 by Supertone, whose model weights are licensed under the OpenRAIL-M licence (full text). Its use restrictions (Attachment A) include, in summary:
- Do not use the model or its output in any way that violates applicable law.
- Do not use it to exploit or harm minors, or to generate or spread verifiably false information meant to harm others.
- Do not use it to generate personal data that could harm an individual, or to defame, disparage or harass others.
- Do not publish machine-generated content without clearly disclosing that it is machine generated.
- Do not use it to impersonate others (including deepfakes) without their consent.
- Do not use it for fully automated decisions that adversely affect someone’s legal rights, or to discriminate against individuals or groups.
- Do not use it to exploit the vulnerabilities of a specific group, to give medical advice, or for justice, law-enforcement, immigration or asylum decisions.