Direct the performance, keep a long book consistent, and prove it is right before you export — including the check that refuses an export when words are missing.
Lesson 3 · core · 9 min read
Open Audiobook StudioA book is long enough that consistency beats perfection. The goal of a review pass is not a flawless take of every paragraph — it is a performance that holds the same character for eight hours, with nothing in it that a listener would stop and rewind.
Everything in the studio applies in layers: the book-level settings underneath, per-line settings on top. Start at the bottom of that stack, because one change there moves every line and costs you nothing to try.
The controls that shape a performance, and where each one lives.
Tip
One paragraph describing how the whole book should be performed. It steers the per-line direction the studio suggests, so it is the cheapest way to move every line at once. Free text, up to 600 characters. Leave it empty and the studio writes one from your manuscript.
The style presets are Narrative, Conversational, Warm, Dramatic, Calm and Energetic, and speed runs 0.8× to 1.2×, in steps of 0.05. Both are project-wide, and both take effect on the next render rather than on audio you already have.
Once the book-level performance is right, most paragraphs need nothing. Reach for a per-line control where the text is genuinely ambiguous — a line that is sarcastic on the page, a whispered aside, a shout — and leave the rest alone.
Overrides the narrator for this paragraph only — the control behind a second character reading their own dialogue. Options: Any voice in the picker.
The feeling behind this line. One choice, applied to the whole paragraph. Options: neutral, warm, tense, sad, excited, intimate, dramatic, authoritative.
A short performance note for this line, in your own words — the audiobook equivalent of a margin note to a narrator. Options: Free text, e.g. “Low, urgent whisper”.
How much silence follows this paragraph. Auto lets the engine shorten the gap between lines from the same speaker; the rest are fixed holds. Options: Auto, 0 ms, 150 ms, 250 ms, 500 ms, 1 s, 2 s.
Note
Every per-line control is an instruction for the NEXT render. Changing one on a paragraph that is already narrated does not alter the audio you have — and when it is the direction you changed, the studio badges the row so the mismatch is visible rather than something you discover in the export.
The three casting actions, all in the Cast sheet.
Cast in that order. Analyze script is what turns a wall of narration into narrator lines and character lines; without it there is nothing for the other two to work on. A character called by more than one name — a nickname, an initial, a title — can be pointed at a single canonical voice, so the same person does not arrive in three different voices.
Proof by listening, with the text in front of you. The player follows the paragraph rows, so you can hear a line and mark it without losing your place. Flag first and fix in a second pass — stopping to fix each line as you hear it turns a chapter into an afternoon.
Two badges appear without you asking for them, written by the studio rather than by you. They are the ones worth understanding, because one of them can stop an export.
Badges the studio puts on a row by itself.
Careful
“May be missing words” is a hard gate, not a suggestion: the export refuses while any paragraph is flagged, until you re-narrate it or explicitly export anyway. It exists because a missing sentence is the one defect a listener always notices and an author almost never does.
The export sheet runs a readiness checklist first. It separates what a retailer will reject from what is merely optional, so you are not guessing which warning matters.
The readiness checklist, and what actually blocks a submission.
Then choose what to package. MP3 — ACX-spec (192 kbps) and WAV — lossless master — the first is the submission package, the second is a lossless master for archiving or for editing elsewhere.
The optional extras on a submission package.
Room tone is handled for you. Every exported chapter is padded with 0.9 seconds of silence at the head and 3 at the tail — inside the 0.5–1 second opening and 1–5 second closing window a retailer expects — and you can change both if you have a reason to. Loudness is normalized during the export, so that is one spec you never have to think about.
Tip
Cut the retail sample from a chapter that shows the book at its best rather than from chapter one out of habit. The lengths on offer are 1 minute, 2 minutes, 3 minutes and 5 minutes, and it is the only part of the book most browsers will ever hear.
A book that clears the checklist, has no outstanding “May be missing words” flags and sounds like one narrator throughout is a book you can submit. The cover is the item people leave to last and the one a retailer rejects first — no amount of good narration substitutes for it.
Questions