Save spoken narration inside an EPUB with calibre

An EPUB can carry spoken narration as well as text. Calibre’s editor can generate that speech and package the audio with timing information that links it to passages in the book. A compatible reader can then play the stored recording while you follow the words. The speech is generated locally.[4]

That is useful when you want a particular narration to travel with a document, including to reading software that supports audio overlays but has no speech generator of its own. The narration becomes part of the EPUB. Changing a reader’s live text-to-speech voice does not remake that saved recording.[3][4]

The instructions below were checked against the official documentation and calibre 9.16.0 source code; no speech-generation run was performed for this guide. The Windows download page lists 9.16.0 and requires Windows 10 version 1809 or later. Start with that desktop release and a short, editable EPUB containing selectable text.[1]

Prepare the book before adding speech

Keep a separate copy of the original EPUB. Open the working book in calibre, right-click its library entry and choose Edit book. In the editor, choose Edit, then Create checkpoint, so there is a named state to return to during this editing session. The file browser on the left shows the book’s contents.[2]

A plain EPUB is a useful starting point because this narration command supports EPUB-family books. If its internal package is older than EPUB 3, the generation routine upgrades it. The routine also removes existing audio overlays before building new ones, so use a copy if the book already contains a recording you want to retain.[4]

Choose the voice, then generate

In the editor’s Tools menu, choose Add Text-to-speech narration. Under Choose default voice for language, expand the relevant language and tick a voice. The list allows one preferred voice per language. The dialog also offers Speed of speech; for a first attempt, leaving the existing speed setting gives you one fewer choice to judge.[3][5][6]

Choose OK to begin. Calibre obtains any voice files it needs before generating the narration. The dialog then reports progress. Its built-in help warns that generation can be slow and depends on the computer’s hardware, which is why a short book makes a better first job than a long novel.[6]

The output includes M4A audio files and SMIL files that map pieces of that audio to the text. Calibre adds those files to the EPUB’s package and records the timing information. You do not have to assemble a separate audiobook or manually align each sentence for this workflow.[4]

Save and check the stored result

After generation completes, choose Tools, then Check book. Review any errors it reports, then use File, Save. Check book inspects the ebook’s structure and links; listening remains a separate check on pronunciation and whether the spoken passages match the text.[2]

Close the editor and reopen the saved EPUB in calibre’s E-book viewer. Go to a passage in the main text. Press Ctrl+S in the viewer, or show its controls and select Read aloud. In version 9.16.0, that action opens the embedded-audio player when the book contains a SMIL overlay; otherwise it opens live text-to-speech.[7][8]

For a useful acceptance check, listen at the beginning of a passage and again farther into the book. Check names, numbers and a sentence that crosses punctuation. Look for the highlighted text to stay with the recording. Reopening the saved EPUB in the editor and finding its added audio and SMIL files gives a second check that narration was packaged with the text.[4][7]

Keep the narration matched to the text

If you revise the words later, regenerate the narration and repeat the playback check. This follows from how the feature saves recorded speech and timing references: the old audio cannot acquire your new wording. Likewise, choose another voice in the editor and regenerate if you want a different saved narrator.[4][5]

The stopping point is a saved EPUB that opens, plays its embedded audio and follows the correct passages. Test that same file in any other reading app you intend to use, because the receiving app must support audio overlays.[3][6]

Report an error or suggest a correction