AI Audiobooks for ACX, KDP and Other Stores: Rules and Specs

You can publish an AI-narrated audiobook, but not everywhere. At the time of writing, ACX (which supplies Audible, Amazon and Apple) requires human narration unless Amazon authorizes otherwise, and Amazon's own route for AI narration is Virtual Voice, a beta for eligible English Kindle ebooks. Google Play Books, Kobo and Spotify (through Findaway Voices) accept AI narration as long as you disclose it. Store rules change often, so check each store's current guidelines before you upload.

Whichever store you choose, the audio has to meet technical specs. Here's what they are and how to produce a book that passes them.

Which stores accept AI narration?

What are ACX's audio requirements?

ACX's specs are the strictest in the industry, so most authors produce to them even when publishing elsewhere:

• Each chapter as a separate file, plus opening and closing credits and a retail sample.
• RMS between −23 dB and −18 dB, peaks no higher than −3 dB, and a noise floor no higher than −60 dB.
• MP3 at 192 kbps or higher, constant bit rate, 44.1 kHz.
• 0.5 to 1 second of room tone at the start of each file and 1 to 5 seconds at the end.

AI narration has an easy time with the noise floor (there's no room noise), so the main work is loudness and file structure.

How do I make an audiobook in Leelo?

1. Start an Audiobook project.
2. Bring in the manuscript with Paste a document: Word (.docx), EPUB, PDF, .txt or Markdown. With "headings start chapters" on, each heading becomes a chapter. PDFs come in without their headings, so a Word or EPUB file saves you work.
3. Choose a narrator, and give characters their own voices if you want a full-cast feel.
4. Read a chapter at a time, or the whole book, and fix names in the pronunciation dictionary.
5. Add opening and closing Credits and a retail Sample.
6. Export every chapter as its own MP3 (at the Audiobook/ACX loudness of −20 LUFS) or one M4B with chapters. The export runs an ACX check before you download.

Will listeners notice it's AI?

Over hours, a monotone voice gets tiring, so long-form needs more direction than a short video: vary the pace, add pauses at scene breaks, and use a different voice for dialogue if your book has a lot of it. Nonfiction, guides and self-help tend to suit AI narration best; dialogue-heavy fiction needs the most care.

What does it cost?

A typical 70,000-word novel is about 70,000 credits in a standard voice (a credit a word) or 105,000 in a premium voice. The Starter plan includes 80,000 credits a month; for bigger books, use a larger plan or a credit pack. The cost shows before you generate.

More on the workflow: audiobooks with AI voices.