An 80,000-character audiobook, overnight

Listen instead · voiced on a MacBook Air, offline

The audiobook was invented as an accessibility project, and it began as a logistics problem. In 1934 the American Foundation for the Blind and the Library of Congress started pressing novels onto long-playing records at thirty-three and a third revolutions per minute, a speed chosen years before the format reached music, because it squeezed roughly fifteen minutes onto a side. Fifteen minutes. A single ordinary novel ran to twenty or more discs, packed in a case heavy enough that the postal service wrote special rules for shipping them, which is why library books for the blind still travel free in the United States today.

Everything about making one was slow. A narrator read into a lathe cutting a lacquer master in real time, and a fluffed line near the end of a side meant cutting the side again from the top. The reason audiobooks stayed rare for fifty years was not lack of demand. It was that each one cost a studio, a union narrator and several weeks.

You can now start one at eleven at night, on a laptop you already own, and find it finished at breakfast. Here is how to do that in a way that produces something worth listening to, rather than nine hours of technically correct audio nobody can sit through.

Bring the manuscript in whole

Drop in the file as it exists: .epub, .pdf, .docx or .txt. Do not pre-split it into chapter files, because the chapter structure is the thing everything else hangs off, and it is better detected from the whole document than reconstructed from filenames.

What you get back is a proposed chapter list, and this is the screen worth slowing down on. Detection is good and it is not infallible, particularly with PDFs, where a running header repeated on every page can masquerade as a heading and a genuine chapter opening set in small caps can be missed entirely. Read the list. Fix the titles. A wrong chapter break is cheap to correct here and annoying to correct later, because chapters are the unit of everything downstream: they are how the audio is stitched, how it is delivered, how it is played, and what you re-run when you change your mind.

Cast the book before you render it

If your book has dialogue, the difference between a flat read and a real production is who says the quoted lines. Usovo works out which character each quotation belongs to, then hands you the result to review before anything is synthesised, and the review is not a formality. Attribution is genuinely hard in fiction, where a page of back-and-forth can run for a dozen exchanges without a single he said to anchor it.

So scan the list, correct anything obviously wrong, and then cast deliberately. Pick voices that differ in register rather than in accent, because a listener two hours in and half asleep distinguishes a low voice from a high one long after they have stopped noticing whose vowels are whose. Give the narrator the voice you would happily hear for nine hours, not the most characterful one on the list. If none of the presets fit, a voice you cloned yourself works here exactly as it does anywhere else. Characterful gets tiring in a way that is impossible to detect in a thirty-second audition and obvious by chapter four.

The queue is built on the assumption that long jobs get interrupted, because they do. Chapters are tracked individually, and a job that is stopped by a closed lid, a crash or a flat battery picks up from the chapter it was on rather than starting the book again. Nothing about a nine-hour render should require the machine to have a perfect night.

The people pressing lacquer masters in 1934 would have found the speed of this ridiculous, but they would have recognised the method immediately. Get the structure right first. Cast it properly. Listen to a chapter before you commit to the book. The tooling collapsed from weeks to a night, and almost everything that makes an audiobook good survived the collapse unchanged.

· · ·