A book workflow, from import to listening
An AI audiobook starts with a manuscript you can legally use. This workspace is built for an author reviewing a draft, an educator preparing permitted lesson material, or a reader making an accessible listening copy of their own text. Begin with a small chapter so you can check the extraction and narration before committing to a larger batch. A short preview reveals names, punctuation and pacing problems earlier than a full-book render.
The project preserves the book’s structure rather than summarizing its ideas. Import the whole supported document, review its chapters, and decide what belongs in the narration. You can exclude a chapter without deleting its words. Read the original alongside your editable narration text, then save the reviewed manuscript to your private account. The original upload is parsed locally in your browser; saving sends the extracted text, not the binary book file.
Choose the file that preserves your chapters
EPUB is a useful starting point when the ebook already has a clean reading order. The importer follows its spine and treats section headings as chapter boundaries. Navigation-only files are skipped. Encrypted EPUB resources are rejected, and this tool does not remove DRM. Compare the result with your ebook: a title page or appendix can still be readable content even when you do not want it spoken.
DOCX works best with real Heading 1 or Heading 2 styles and ordinary body paragraphs. A large bold line alone may not be recognized as a heading. Tables, tracked revisions, text boxes, images and complex footnotes need particular care; do not assume that a visually correct Word document has become a complete narration script. TXT and pasted text use blank lines for paragraph boundaries and recognizable chapter headings for section breaks. Save TXT as UTF-8. PDF, scanned pages and OCR are outside this version.
Chapters organize; passages generate
A chapter is the unit for organization and listening. A passage is the unit for narration and correction. Rename or reorder chapters, exclude front matter, and adjust a passage without rewriting the source. A very long paragraph is split into bounded passages so that a single provider request stays within its input limit. You can also split at the editing cursor or start a new chapter at a passage boundary.
Choose one narrator, the text language and a comfortable speech speed. Preview a real passage from the book; the preview is an actual generation and carries the same credit rate. When the voice and text are ready, generate the missing audio for a chapter. The service reuses unexpired audio that matches the exact passage text, narrator, language and speed. Editing one passage invalidates that passage’s match. Choosing another narrator or speed requires new audio for every passage affected.
Review what an AI narrator actually says
Listen for names, abbreviations, numbers and words with more than one pronunciation. Sentence breaks influence how a narrator pauses, but the final delivery depends on the selected model. A passage preview helps you compare your intended meaning with the generated speech. No voice setting guarantees a particular emotional performance, and a successful API response does not establish publication-quality narration.
Keep corrections purposeful. Expand an abbreviation in the narration text, adjust punctuation, or restore the extracted original if an edit changes the meaning. Proofread any content that must stay faithful to an ebook edition. This version uses a single narrator throughout a generation batch; it does not detect fictional characters, assign dialogue voices, clone a person, add music or create a dramatized performance. Provider voices are presets. Confirm the provider’s current usage terms before using generated speech commercially.
Know the cost before you generate
Generation costs one credit for every started 100 characters in each selected passage. Rounding happens separately for each passage, so many short paragraphs can cost more than the same text in a longer passage. The server calculates the quote from your saved manuscript. It shows the passage count, total characters, credit cost and balance before you confirm. Quotes expire after five minutes and are bound to the saved book version.
Only one narration batch can be active per account. You can leave the page and return to its saved status. Completed passages are archived privately as the job advances. If the batch fails or its outcome remains uncertain, the batch credits are refunded once. A lost provider receipt is not automatically sent again, because that could create a second paid request. Reopening the page is different from choosing to generate a new version.
Download the audio you have reviewed
Play an available passage or listen to a complete chapter in order. Download an individual MP3, or export a chapter ZIP containing numbered passage MP3 files and its narration text. ZIP export requires current audio for every included nonempty passage and is limited to 64 MB of audio. It packages existing files only; downloading never starts synthesis or silently spends credits.
Audio is kept privately for seven days from the job’s creation, so download the files you want to keep before they expire. Saved manuscripts stay in your account until you delete them. This version does not merge a chapter into one MP3 or produce M4B files. A downloaded archive is not a distribution submission: audiobook stores have their own formatting, rights and AI-narration policies. Check the intended store’s current rules before planning publication.