Documentation contents

Documentation

Book sources and cleanup

Supported documents, PDF extraction, headings and material to leave out.

  • New users
  • Operators
flowchart LR
  A[Import a book] --> B[Extract text]
  B --> C[Review cleanup]
  C --> D[Select cleaned source]
  D --> E[Prepare speech]
Extraction, cleanup and speech preparation create separate results.

Input formats

TXT, PDF, EPUB, DOCX, MOBI and pasted text are supported. DOCX and MOBI conversion require Calibre. A clean EPUB or plain-text source can be easier to prepare than a PDF made of scanned pages.

PDF extraction includes OCR options for pages without usable text. Check columns, hyphenated line endings, page numbers, repeated headers, footnotes and tables. OCR can produce plausible-looking errors, so inspect the actual extracted text.

Cleanup decisions

Keep headings that should become chapters. Decide whether to read footnotes, citations, contents pages and publishing information. Inspect the proposed cleanup before continuing; remove repeated furniture without deleting the book's content.

Cleanup is separate from narration segmentation and speech optimization. Book translation and document correction are not dedicated audiobook stages in this release. Prepare a translated or corrected source externally when needed, then import it.

Review extraction before cleanup

Deterministic extraction runs first and keeps an extracted-text baseline. Optional model-assisted cleanup builds on that result. Compare the proposed changes with the baseline when a page is missing, columns are interleaved or headings look wrong.

For a scanned PDF, check OCR language and a page with difficult layout before processing the whole book. For EPUB, inspect a chapter boundary, a footnote and a section with dialogue. Conversion to plain text cannot preserve every feature of a designed page.

Decide what should be spoken

  • Keep chapter headings you want to hear and use for navigation.
  • Remove repeated page furniture after checking that it is actually repeated furniture.
  • Decide whether footnotes and citations should be read, retained as ordinary text or omitted.
  • Check front matter and contents pages separately from the main book.
  • Keep the extracted baseline and the selected cleaned result distinguishable.

After cleanup, select the cleaned source before preparing the speech plan. If you later choose another source revision, check the prepared passages and generated audio against it before reusing them.

See importing and cleaning a book.