Your Audiobook and Your Ebook Can Be One Product: Build a Read-Along Edition With AI Narration (EPUB 3 Media Overlays, 2026)

1. What a read-along ebook actually is

In EPUB 3, a Media Overlay is a small file written in SMIL (Synchronized Multimedia Integration Language) that links pieces of the book's text to clips of an audio file. Each entry says, in effect, "the element with this ID is spoken from 3.20 to 5.85 seconds in this audio file". A reading app that supports Media Overlays plays the audio and highlights each element as it's spoken. The current specification is part of the W3C's EPUB 3.3 Recommendation.

Apple describes the result clearly for its own store: in Read Aloud books "an audio file using a narrator's voice can read the text on a page and the words can be highlighted as they are spoken", with pages turning automatically if the reader wants (Apple Books Asset Guide).

The unit of highlighting is your choice. It can be a word, a phrase or a sentence, depending on how finely you mark up the text and time the audio. That choice matters more than people expect, and we'll come back to it.

This is different from the audiobook-and-ebook pairing some retailers sell, where a separately published audiobook is synced to a separately published ebook. A read-along EPUB is one file containing both the text and the narration.

2. Where read-along works in 2026

This is the part nobody puts in one place. Check each store's current publisher documentation before you build, because support changes.

Two practical conclusions. First, build fixed-layout. That's where read-along support actually exists, and it suits the books that benefit most (illustrated, early-reader and educational titles). Second, plan Apple Books as the primary store, with Kobo's apps second and direct or institutional sales third. If your audience is mainly on Kindle, a read-along EPUB isn't the right product. Make an audiobook instead.

3. Which books suit it, and which don't

The pattern: read-along suits books where the reader is learning to read the text, not just enjoying the story. That's also where AI narration's weakness (performance) matters least and its strength (consistent, clear reading you can re-render) matters most. For language learners specifically, our listening-practice guide covers pace and level.

4. Why AI narration makes syncing easier

With a human narrator, you record a long take and then work out afterwards where each word falls. That's called forced alignment, and it's why read-along production used to be slow. Open-source tools such as aeneas were built to automate it, producing SMIL files for EPUB 3 from text and audio.

With AI narration you can work the other way round. Render each sentence or phrase as its own clip. You then know exactly where every clip starts and ends, because you made them. Sentence-level sync becomes a matter of adding up clip lengths, with no guessing.

Word-level highlighting is harder. Rendering each word separately sounds choppy and robotic, so don't do it. Instead, render naturally by sentence and run forced alignment within each short clip to find word boundaries. Alignment on a three-second clip of clean synthetic speech is far more reliable than on a forty-minute human recording.

5. The build, step by step

1. Start from a fixed-layout EPUB 3. If your book is currently reflowable or print-only, this is the main design job. Each page is a separate XHTML document.
2. Give every unit of text an ID. Wrap each sentence, phrase or word you want to highlight in a span with a unique id.
3. Write the narration script from the same text. Export the text in reading order, one unit per line, keeping the IDs alongside. Fix pronunciations now: invented names, numbers and abbreviations (our normalisation guide helps).
4. Render per unit. Generate each sentence or phrase as its own clip with the same voice and settings throughout. Keep the files named by ID.
5. Join clips per page into one audio file, recording each clip's start and end time as you go. Add a short, consistent pause between sentences and a slightly longer one at the end of the page.
6. Write the SMIL file for each page: one par element per unit, pairing the text ID with the audio file and its clip times. Many EPUB tools can generate this from a timing list.
7. Declare the overlays in the package file. Each page's manifest item gets a media-overlay attribute pointing to its SMIL file, and the package metadata includes the total duration and the CSS class used for the highlight.
8. Style the highlight. Define the active class in your book's CSS. Make it readable for children and low-vision readers: a background tint usually reads better than a colour change on the text alone.
9. Validate. Run EPUBCheck, then test on real devices (section 7).

Audio formats: retailers generally expect AAC (.m4a) or MP3 for embedded audio. Check each store's current spec for codec and bitrate before encoding the final files.

6. Narration choices that matter for read-along

• Slower than an audiobook. The listener is also reading. A pace that sounds natural in an audiobook leaves an early reader's eye behind. Render a sample page at two speeds and test it with a reader from your audience.
• Even emphasis. Heavy dramatic stress makes some words harder to track. For early readers, clarity beats expressiveness.
• Exact match to the text. The narration must say exactly what's on the page. No added "she said", no skipped words, no expanded abbreviations the page doesn't show. A mismatch breaks the highlight and confuses a child learning to read.
• One voice, held constant. If you re-render a page later, use the same voice and settings so it matches its neighbours.
• Two versions for language learners. A slow version and a natural-speed version of the same book is a genuinely useful product, and with AI narration it's a second render, not a second recording session.

Read-along books are also an accessibility feature: synchronised text and audio help many readers with print disabilities. If you sell to schools or public bodies, that's a selling point, and our European Accessibility Act guide covers the wider requirements.

7. Testing before you publish

1. Every page, every device you're selling to. At minimum an iPad in Apple Books and an Android phone in the Kobo app.
2. Watch the highlight, not just the audio. Does it land on the right word? Does it drift later in the page?
3. Test page turns. With auto-turn on, does the next page start cleanly?
4. Test a mid-page start. Tap a word in the middle of a page and check playback starts there, if the app supports it.
5. Listen with a real reader. One session with a child or learner from your audience tells you more about pace than any setting.

8. FAQ

An ebook with built-in narration where words or phrases are highlighted as they're read aloud. In EPUB 3 this uses Media Overlays: SMIL files that link text elements to clips of audio.

Yes, as Read Aloud books, but only in fixed-layout format. Apple also requires a custom preview for Read Aloud titles.

Not as an EPUB 3 Media Overlays book through standard KDP publishing. Amazon's text-and-audio experience pairs an ebook with a separately published audiobook.

Yes, and it suits the format well for early readers, language learners and non-fiction, where clear reading matters more than performance. Check each store's current policy on AI-generated audio before publishing.

Words for early readers and beginners, sentences or phrases for everyone else. Render by sentence for natural audio, then align words inside each short clip if you need word-level highlighting.

You need a fixed-layout EPUB 3, a way to produce timings (clip boundaries or a forced-alignment tool), and the SMIL files. Several EPUB tools can generate SMIL from a timing list. Always validate with EPUBCheck.