How to Narrate Long Text with AI Voices

To narrate long text with AI, give each chapter its own script and each paragraph its own line. abcVoiceover makes a long line a few sentences at a time and joins the pieces seamlessly, so lines can be up to 2,000 characters, but paragraph-sized lines are easier to review, re-take and edit, and they keep the narration steady over hours of audio.

How long lines are made

A line is split at sentence endings into pieces of about 20 seconds of speech each, made one after another in the same voice, and joined with a short pause about the length of a sentence break. Each piece is checked with speech recognition and made again if it skips words or runs on. The result is one take for the whole line, however long it is.

The limit is 2,000 characters a line, roughly 300 to 350 English words or a little over two minutes of speech.

Why a paragraph per line works best

Re-takes are small. If one word is wrong, you re-generate a paragraph, not a chapter.
Review is easier. You can listen paragraph by paragraph, with the text right there.
Files line up with the text. The download numbers every line, so 027 Narrator.wav is paragraph 27.
Pauses are yours. The gap between paragraphs is set when you join the files, so you can give a scene break more room.

Setting up a long project

One project for the book, course or series, with the narrator in its cast.
One script per chapter. Scripts come out in the download as numbered folders, in the order they were made, so create them in reading order.
Paste the chapter with a blank line between paragraphs; each paragraph becomes a line, and blank lines are skipped.
Give headings their own line (Chapter Four. The Harbour.) so they can have a pause after them.
Dialogue in other voices, if you want a full cast: split quoted speech onto its own lines with the character's name. See formatting dialogue.

Keeping hours of narration even

Use the Calm or Balanced style and keep expressiveness moderate, so paragraphs sound alike.
Don't change the narrator's sample or settings partway through.
Keep paragraphs roughly similar in length; split very long ones.
Fast draft is handy for hearing a chapter's flow in English quickly, but it uses a different model. Re-generate final takes in the main style.

Reviewing a long recording

An hour of narration has hundreds of chances for a slip. Review in a quiet place, on headphones, reading along:

Generate the chapter with Generate all new lines, and let it run in the background.
Listen through once, line by line. Save the lines that are right as you go.
For the rest, note the problem: a wrong word, a strange emphasis, a rushed sentence.
Fix the text where the text is the problem, then generate another take of each line you flagged.
Listen to the chapter in sequence once more before you export, to catch anything that sounds out of place next to its neighbours.

If someone else is proofing, share the project with them so they can listen and leave notes in its comments.

Finishing

Download the project, join each chapter's numbered files in an audio editor with consistent gaps, and master them to your platform's loudness target. See editing in Audacity and loudness standards.

How to narrate long text with AI voices