How a book becomes a full-cast audiobook
No one ever steps into a recording booth. Here is the whole journey a novel takes — 31 times over, 159 hours of it.
A clean copy of the book
Every recording starts from a professionally proofread public-domain edition, mostly from Standard Ebooks. This is not housekeeping: a scan with stray page numbers and broken paragraphs derails every step that follows, because the pipeline reads structure as well as words.
Casting the book
A language model reads the novel the way a casting director would — who speaks, how often, and what are they like? Confident or timid, young or old, warm or cold. Every character with a speaking part is matched to a distinct voice, and no two characters in one book are allowed to sound alike.
Working out who is talking
A second model goes through each chapter paragraph by paragraph and decides, line by line, whether this is the narrator describing a room or a specific character speaking, and in what emotional register. This is the step that lets one paragraph move between narration and two people arguing.
Performing the narration
Each book has one narrator, chosen to match its tone — a measured voice for a Gothic novel, a brighter one for an adventure. The narrator is generated by a fast text-to-speech model, which is why 159 hours of audio was affordable to produce at all.
Performing the dialogue
Every character’s lines are generated in the voice cast for them, from a short reference clip in an openly licensed speech corpus. The same reference produces a different performance depending on the mood of the line: afraid, sarcastic, grieving, delighted.
Assembling the chapter
The pieces are joined with the pauses a listener expects — half a second when the speaker changes, slightly less between sentences by the same speaker — then levelled so nothing is inaudible and nothing is shouted, and encoded as one file per chapter.
Why it only works on some books
A full cast earns its keep in proportion to how much of a book is people talking. In a Sherlock Holmes story, where whole pages are dialogue between three or four people, hearing each of them separately removes work the listener was doing without noticing. In a book that is mostly interior narration the same production has almost nothing to do — which is why this catalogue is dialogue-heavy fiction and not simply whatever is out of copyright.