How to Turn Classic Literature into Audiobooks with AI Voices

You can turn a public-domain classic into a good audiobook with text-to-speech, as long as you prepare the text first. The voice is rarely the problem. What trips up AI narration in older books is archaic spelling, character names, Roman numerals, verse meter and long sentences full of dashes, and each of those can be fixed before you generate a single line.

This guide covers the whole job: checking the book is free to use, cleaning the text, fixing pronunciation, handling dialogue, producing chapter files and estimating cost.

Which classic books can you legally narrate?

Start here, because a beautiful recording of a book you don't have rights to is a problem.

Public domain depends on the country

Where you publish	Rule of thumb in 2026
United States	Works published in 1930 or earlier are in the public domain.
UK and EU	Works are generally free 70 years after the author's death, so authors who died in 1955 or earlier.
Elsewhere	Terms vary. Check the local rule before you publish there.

So Pride and Prejudice (1813), Moby-Dick (1851) and Jane Eyre (1847) are free nearly everywhere. A book published in the 1930s by an author who died in the 1960s may be free in the US but not in Europe.

Watch out for translations and editions

• Translations have their own copyright. Dostoevsky is in the public domain, but a recent translation of Crime and Punishment isn't. Older translations, such as Constance Garnett's, often are.
• Modern editions add new material. Prefaces, footnotes and editorial corrections can be protected even when the novel isn't.
• Project Gutenberg texts are a common source. The site notes that its books are public domain in the United States, and readers elsewhere should check their own country's law. Remove the Gutenberg header and licence text before you narrate.

This is general information, not legal advice.

How do you prepare a classic text for text-to-speech?

Most narration problems are text problems. Spend an hour here and you'll save many hours of regenerating lines.

Clean the source

• Delete page numbers, running headers, illustration captions and transcriber's notes.
• Join lines that were broken mid-sentence by the original layout.
• Replace underscores or asterisks used for italics. Decide whether each emphasis should be spoken with stress or ignored.
• Move footnotes out of the text, or drop them. A footnote read in the middle of a sentence confuses listeners.

Expand what the voice can't guess

In the text	Say it as
Chapter XIV	Chapter Fourteen
Henry V, Louis XIV	Henry the Fifth, Louis the Fourteenth
&c.	et cetera
£5 10s.	five pounds ten shillings
Mr. D—	Mister D (or a short pause)
'tis, o'er, e'er, ne'er	it is, over, ever, never (or keep the contraction and check it)

Then import the cleaned book. Leelo's document import brings the text into the studio as lines you can edit, direct and regenerate one at a time.

How do you get names and archaic words pronounced right?

Classics are full of words a modern voice has never heard said properly. Build a list as you go and put it in the pronunciation dictionary, so every occurrence is fixed at once.

Names

• Hermione in The Winter's Tale: her-MY-oh-nee.
• Gloucester in King Lear: GLOSS-ter, not glow-chester.
• Raskolnikov: stress on the second syllable, ras-KOHL-nee-kov.
• Russian patronymics like "Rodion Romanovich" appear hundreds of times in a long novel. One dictionary entry covers all of them.

Heteronyms

English words spelled the same but said differently trip up any voice, human or synthetic:

• "The wind howled" vs "wind the clock"
• "a tear in her eye" vs "tear the letter"
• "she read it yesterday" vs "read it aloud"
• "lead pipe" vs "lead the way"

Listen for these on your first pass. When a line gets one wrong, fix that line rather than the whole chapter.

Verse and meter

Poetry needs extra care. Shakespeare editions often print "belovèd" with a grave accent to show the final syllable is spoken (be-LUV-ed) so the line scans. Older poems sometimes rhyme "wind" with "kind." Decide whether you want historical pronunciation or modern, and be consistent.

Should a classic have one narrator or a full cast?

Both work. The choice depends on the book.

Single narrator

Best for first-person novels (Jane Eyre, Moby-Dick), essays and most 19th-century fiction, where the narrator's voice is the book. Pick one warm, steady voice from the English library with a pace you can listen to for hours.

Narrator plus character voices

Best for plays, books heavy with dialogue, and children's classics. In Leelo each line has a speaker, so you can assign the narrator to description and different voices to characters in the same project:

Narrator: "It is a truth universally acknowledged…" Mrs. Bennet: "My dear Mr. Bennet, have you heard that Netherfield Park is let at last?" Mr. Bennet: "I have not."

Keep the number of character voices small. Three or four main characters in distinct voices is plenty; minor characters can be voiced by the narrator.

Directing the performance

Classic prose has long sentences with semicolons and dashes. Add pauses at the dashes, slow down for important passages, and keep several takes of key lines (the closing line of a chapter, a famous quotation) so you can choose the best one.

How long will the audiobook be, and what will it cost?

Audiobook narration usually runs at about 150 to 160 words a minute. A worked example for Pride and Prejudice, which is roughly 120,000 words:

• Running time: 120,000 ÷ 150 is about 800 minutes, or roughly 13 hours.
• Credits with a standard voice: about 120,000 (1 credit a word).
• Credits with a premium voice: about 180,000 (1.5 credits a word).
• Allowance for fixes: add 10–20% for regenerated lines and alternate takes.

The Creator plan ($39 a month, 200,000 credits) covers the whole novel in a premium voice in one month. On Starter ($19, 80,000 credits) you'd spread it over two months or top up with a credit pack, which lasts 12 months. Failed jobs return their credits. See the pricing page for every plan. Commercial use of the audio, including selling the audiobook, is allowed on paid plans.

How do you export an audiobook that meets platform specs?

Retailers and distributors publish technical requirements. The widely used specs that ACX publishes for audiobooks are a good baseline even if you sell elsewhere:

• One file per chapter or section, each under 120 minutes.
• Opening credits (title, author, narrator) at the start and closing credits at the end.
• Half a second to a second of room tone at the start of each file and one to five seconds at the end.
• Loudness between -23 and -18 dB RMS, peaks no higher than -3 dB, noise floor below -60 dB.
• MP3 at 192 kbps or higher, 44.1 kHz, constant bit rate.

Export each chapter separately (MP3 or WAV) from Leelo, then check a few files with the free loudness checker. Platforms also set their own rules on AI-narrated books, including disclosure, and these change. Check the current policy of the store you plan to use before you upload.

The audiobook help guide walks through a full project, and the audiobooks use case page has examples.

Who benefits from audio versions of the classics?

• Students working through set texts, especially those who find dense 19th-century prose hard to read. Listening while following the printed text can help, as we explain in text-to-speech for dyslexia.
• People with low vision who want a specific edition or translation that has no commercial recording.
• Teachers who want an audio version of one chapter or one poem for a lesson, read the way they want it read.
• Small publishers reissuing out-of-print public-domain titles with a new audio edition.
• Commuters and walkers who'd never sit down with Middlemarch but will listen to it.

Frequently asked questions

Is it legal to make an audiobook of a public-domain book?

Yes, if the text you're using is genuinely public domain where you publish. Check the publication date and the author's death date, and make sure you aren't using a modern translation or edition with its own copyright. Your new recording is your own work.

Can I sell an AI-narrated audiobook of a classic?

Leelo allows commercial use of the audio on paid plans. Whether a particular store accepts AI narration, and how it must be labelled, depends on that store's current policy. Many classics already have audio editions, so a careful text, good casting and a clear listing matter.

How do I stop the AI voice mispronouncing old words and names?

Add each one to the pronunciation dictionary once, and it applies to every line in the project. For heteronyms like "wind" or "tear," fix the individual line. A first full listen-through usually catches most issues.

How long does it take to produce a full novel?

Generation is fast; preparation and review take the time. Expect to spend a few hours cleaning the text and building the pronunciation list, and then real-time listening for quality control: about 13 hours for a 120,000-word novel if you check every line. Spot-checking is faster but riskier.

Should I use one voice or several for a classic novel?

One narrator suits most novels, especially first-person ones. A narrator plus a few character voices works well for plays, dialogue-heavy books and children's classics. Use distinct voices for the main characters only and let the narrator handle the rest.