Audiobooks

flowchart LR
  A[Book or text] --> B[Extract and review]
  B --> C[Speech plan and voices]
  C --> D[Generate a sample]
  D --> E[Generate remaining speech]
  E --> F[Assemble and export]

From book to speech

Import a source and review the extracted text. Cleanup can remove repeated headers, tables of contents and other material you do not want spoken. Segmentation prepares manageable narration units. Speech preparation can expand abbreviations and add reviewed pronunciations.

Source cleanup and speech preparation have different purposes. Keep normal written text for reading or video-book captions, and use the speech text for pronunciation changes.

Generation and review

Choose a service, model, voice and language. Audition a short passage. Generate speech, listen to the takes and replace individual problem segments. Assemble a selected generation run before exporting.

Multiple voices use a cast and speaker assignments. Their requests are generated separately and assembled into the project. See multiple voices.

Output

M4B supports audiobook metadata, a cover and chapters. MP3, Opus, FLAC and WAV are also available. A video book adds displayed passages to narration. Check chapter order, title, author and cover in a real player after exporting.

Before a long generation

1. Read the beginning of the extracted text and a passage from the middle. Check that chapter headings, dialogue and footnotes are in the intended order.
2. Prepare the speech plan from the cleaned source you selected. Start with model-aware chunking; a suitable passage length depends on the speech model.
3. Choose one narrator or assign the cast. Generate a paragraph containing a name, a number and any dialogue you expect to repeat.
4. Listen before generating the remaining passages. If you change the voice, model or pronunciation settings, check which takes need regeneration.

Replacing a passage

A new take and an assembled audiobook are separate results. After fixing a passage, assemble the intended run again and check its joins with the preceding and following passages. A later run can retain unchanged takes from earlier work; it does not necessarily contain newly synthesized audio for every row.

Choosing an output

Chapter markers are embedded in M4B. A WAV export does not carry the same audiobook tags or cover. Use a JPEG or PNG cover for a release-0.12.0 M4B, and check it in the target player. Keep the source project when you want to revise the book later.

See the audiobook tutorial and the narration examples.