Preparing your manuscript for narration

The narrator gets the text exactly as you hand it over. Ten minutes spent tidying a manuscript will do more for the finished audiobook than any amount of re-rendering afterwards, because most of what sounds wrong in a bad narration was already wrong on the page.

Here is what happens to your file, and what to do about it.

What the uploader actually reads

You can paste text straight in, or upload a PDF, DOCX, TXT or Markdown file up to 10 MB. Uploaded files are parsed in your own browser — the text is pulled out on your machine and shown to you before anything is sent anywhere. Read that extracted text. It is what will be narrated, and it is the last easy moment to fix it.

Blank lines are the instructions

Your book is split into narration-sized chunks before any audio is made. The split walks the text paragraph by paragraph, packs paragraphs together until a chunk is around 750 characters — roughly a printed page — and stops well before 1,100. It never cuts a paragraph in half. A single paragraph longer than the ceiling is the only case where the text is broken up internally, and then it breaks between sentences, never mid-sentence.

Paragraphs are found by blank lines. That one detail decides how your book is cut:

• Keep a blank line between paragraphs. This is the normal output of every word processor, so it usually costs you nothing.
• A manuscript that is one continuous block, or one that uses single line breaks for every new paragraph, gives the splitter nothing to work with and falls back to breaking on those hard line breaks.
• A scene break you want honoured should be a real blank line, not a row of asterisks squeezed against the text above it.

Cut the page furniture

A PDF made for print carries things a reader's eye skips and a narrator will not: running heads, page numbers, "Chapter Four" repeated in the margin of every page, footnote markers stranded mid-sentence. All of that becomes speech.

Before you upload, strip:

• headers and footers, page numbers, and any repeated title line
• footnote and endnote markers, unless you want them read
• tables of contents, index pages, and copyright boilerplate
• hard hyphens left over from justified typesetting, which turn "wan- dering" into two words

A scanned PDF has no text in it

This is the failure that surprises people. A PDF produced by a scanner is a stack of photographs of pages. There is no text layer to extract, so extraction comes back with almost nothing and the upload is rejected rather than silently narrating a blank book. Run it through OCR first, or paste the text in from wherever you still have it.

Narrate one chunk before you narrate ninety

Once the text looks right, generate a single chunk and listen to it — a chunk with dialogue in it, ideally, and a name or two you care about. You will hear immediately whether your punctuation is carrying the phrasing, whether an unusual name comes out the way you meant, and whether the voice suits the book.

Credits are spent on the words you actually narrate, so this costs you one chunk. It is the cheapest quality check available to you, and almost everyone who skips it ends up re-rendering more than one page.