How to proof an audiobook

Audiobook proofing is quality control between the narration booth and the listener. Someone listens to the finished recording against the final manuscript and logs every place the two disagree. Here is how the job is actually done, and where the hours go.

What proofing is (and is not)

Proofing is not editing and it is not mastering. The proofer’s job is a list of faults with timestamps, sent back so the narrator can record pickups. The text is the authority. If the book says “melded” and the narrator said “molded”, that is a finding, however natural it sounded.

What you are listening for

• Wrong words. Substitutions that change or blur meaning: “death” read as “debt”, “fight” as “right”, tense slips like “was” for “is”.
• Skipped text. A dropped word, sentence, or in the worst cases a paragraph lost to a bad edit.
• Extra words. Insertions that aren’t in the manuscript, often from a re-take stitched in loosely.
• Repeated takes. False starts and doubled words the edit was supposed to remove.
• Pronunciation. Proper nouns above all. Log the choice the narrator made so the publisher can rule on it once, not per chapter.
• Noise. Mouth clicks, harsh sibilance, breaths left at edit points, room tone changes, long dead air.
• Front and back matter. Credits read exactly to the tag script, correct title, author, narrator.

The arithmetic of doing it by ear

A typical trade book runs ten to fourteen finished hours. Proofed manually, the industry ratio is between 1.2× and 2× the runtime. You cannot skim, because a clean-sounding sentence can still be the wrong sentence. That is one to two full working days per book, mostly spent listening to audio that is perfectly fine, to find the forty places that are not.

Writing a report the studio can use

Every finding needs three things: where (file and timestamp), what was heard, and what the text says. Publishers usually want it in their own sheet format: file number, time, chapter, page, problem. One line per fix, no commentary. The narrator should be able to work down the list cold.

The automated version

This first pass, hearing every word against the manuscript, is exactly what machines are now good at. BookMyProof transcribes each track with per-word timestamps, locates it inside the book, aligns the two word by word, and writes up only the differences that survive a judgment pass (“1,847” read aloud as “eighteen forty-seven” is not an error, and a running head glued into the PDF text is not skipped text). A fourteen-hour book comes back in minutes with each finding one click from its audio. The human stays in the loop for what humans are good at: confirming, dismissing, and judging pronunciation, then printing the sheet.