Seven AI narration tools for audiobooks, ranked: ElevenLabs, Google Play Books, KDP Virtual Voice, Spotify's creation tools, Gemini TTS, WellSaid and open-source Kokoro, with prices, voice quality and which stores accept the result.
ElevenLabs is the best AI text-to-speech for audiobooks: its Eleven v4 voices top the Artificial Analysis speech leaderboard, and its Studio imports an EPUB, detects chapters and casts a different voice for each character.
Google Play Books narrates your book for free, and you can sell the files on other stores as long as the book is also on Google Play.
Audible doesn't accept AI audio you made elsewhere. For authors, the AI route in is Amazon's own KDP Virtual Voice, which is invite-only and US-only.
Gemini TTS costs about $16.50 per million characters, a fraction of ElevenLabs' price for a top-five voice, but it's an API with no editing interface.
PlayHT's service is gone after Meta bought PlayAI in 2025, and Hume's text-to-speech API closes on November 13, 2026, so don't start a long project on either.
Check this before you pick a tool, because the stores set the rules.
Audible and ACX: ACX requires human narration unless it authorizes otherwise, so AI audio you upload is rejected. KDP Virtual Voice, Amazon's own AI narration, is the exception.
Spotify: accepts digital voice narration from Google Play Books, ElevenLabs and others, asks you to tick a digital voice box, and adds a disclosure to the description.
Google Play Books: makes AI narration itself and lets you download the files.
Apple Books: offers its own free digital narration through partners, in English, labeled "Narrated by Apple Books".
Voices by INaudio (formerly Findaway Voices): distributes to most other stores and libraries.
We looked at how natural each voice sounds over hours of listening, long-form tools like chapter import, character voices and pronunciation fixes, which stores accept the result, commercial rights and cost per finished hour.
Tool | Best for | Starting price |
|---|---|---|
ElevenLabs | Best overall | Free (10k credits), Starter $6/mo |
Google Play Books | Free narration | Free (for a limited time) |
KDP Virtual Voice | Getting on Audible | Free (invite-only) |
Spotify Audiobook Creation Tools | Spotify-first releases | Invite-only beta, US authors |
Gemini TTS | Low cost through an API | About $16.50 per 1M characters |
WellSaid | Pronunciation control | Starter $10/mo billed yearly |
Kokoro | Free, local narration | Free (open source) |
ElevenLabs is an AI voice platform, and Eleven v4, released in September 2026, replaced v3 as its flagship model.
What works: Eleven v4 and v4 Turbo hold the top two spots on Artificial Analysis's voice leaderboard, and ElevenLabs says v4 keeps a full audiobook sounding like a single take. In Studio, you import an EPUB or PDF and it detects chapters and casts voices for the characters it finds. Pronunciation dictionaries fix names and invented words, and you can export MP3 or WAV. Text costs one credit per character, so the $99 Pro plan's 600,000 credits cover about 10 hours of audio a month. You can publish to Spotify, Voices by INaudio and its own ElevenReader app, which pays 60% of net revenue.
What doesn't: The free plan has no commercial license, so you need Starter ($6) or above to sell. ElevenLabs doesn't promise ACX-compliant output, and a long book on a top voice costs real money.
Best for: Authors and publishers who want the most natural voices and full control over a production.
Google Play Books turns your ebook into an audiobook for free when you publish through its partner center.
What works: You choose from over 50 voices, assign different voices to characters, and edit pronunciation and pauses. Narration is free for a limited time, and you can download the files and sell them elsewhere as long as the title is also sold on Google Play. Spotify names it as an accepted source, and Google points to Voices by INaudio for wider distribution.
What doesn't: It covers only a handful of languages, including English, Spanish, French, German, Hindi and Brazilian Portuguese, and the voices sound less expressive than ElevenLabs over long dialogue.
Best for: Self-published authors on a budget.
Virtual Voice is Amazon's AI narration for books published through Kindle Direct Publishing.
What works: It's the route for authors to put AI narration of their choice on Audible, and it's free to create. You choose from 80 voices, can change voice by chapter and edit in its Virtual Voice Studio. Books sell on Audible, Amazon and Alexa and in Audible Plus, at $3.99 to $14.99 with a 40% royalty.
What doesn't: It's an invite-only beta, US only, and if your ebook is in KDP Select, the audiobook is exclusive to Audible.
Best for: KDP authors who want their book on Audible without hiring a narrator.
Spotify announced its own audiobook creation tools in May 2026 and opened them as an invite-only beta in June, powered by ElevenLabs voices.
What works: US self-published authors writing in English upload an EPUB, review pronunciations and preview chapters, and most audiobooks go live on Spotify within 72 hours, labeled "Narrated by Digital Voice". You get ElevenLabs voices without an ElevenLabs subscription.
What doesn't: It publishes to Spotify first, with wider distribution through Voices by INaudio, accounts are limited to five projects, and it's US and English only.
Best for: Authors whose readers listen on Spotify.
Gemini TTS is Google's speech model, and its current Gemini 3.8 Flash TTS ranks fifth on Artificial Analysis's leaderboard.
What works: Google pitches it for audiobook and podcast narration, with stable voice and room tone over long passages, and it covers more than 130 languages. It has 30 featured voices plus hundreds more, and it can design a voice from a description. At about $16.50 per million characters on Artificial Analysis's figures, it costs a fraction of ElevenLabs ($40 to $80) and Cartesia ($49).
What doesn't: It's an API, so you'll need a script to feed it chapters and stitch the audio, and one request handles only two speakers. Google says its prices double on January 1, 2027.
Best for: Technical authors and small publishers producing many books.
WellSaid is a text-to-speech studio built for professional voiceover.
What works: Pronunciation controls keep names and terms consistent across a whole series, and you can generate without limits and pay only for minutes you download. Every paid plan includes full commercial rights. Starter costs $10 a month billed yearly for 240 minutes a year, and Pro $33 a month billed yearly for 2,160 minutes.
What doesn't: It's priced per downloaded minute, so a full-length book needs the Pro plan or more, and it has no audiobook-specific tools like EPUB import or character casting.
Best for: Nonfiction authors and publishers who need precise pronunciation.
Kokoro-82M is a small open-source text-to-speech model released under the Apache 2.0 license.
What works: At 82 million parameters it's light enough to run on an ordinary computer for free, and its license allows commercial use. For more expressive voices, Fish Audio's open models rank higher on Artificial Analysis's open-weights leaderboard.
What doesn't: You'll need to be comfortable with Python and splitting chapters yourself, and it has fewer voices and less natural delivery over long dialogue than ElevenLabs.
Best for: Technical authors who want to narrate for free.
Even where ACX won't take AI audio, its specs are a good quality bar for any audiobook. Export each chapter as its own MP3 at 192 kbps or higher, constant bit rate and 44.1 kHz. Keep loudness between -23 and -18 dB RMS, peaks below -3 dB and the noise floor at -60 dB or lower, with 1 to 5 seconds of room tone at the start and end of each file, and no file longer than 120 minutes.
ElevenLabs, for most authors. Its Eleven v4 voices top the Artificial Analysis leaderboard, and its Studio imports an EPUB, detects chapters and assigns character voices. Google Play Books is the best free option.
Only through KDP Virtual Voice, Amazon's own AI narration, which is invite-only and US-only. ACX requires human narration for audio you upload yourself.
It can be. Google Play Books and KDP Virtual Voice narrate for free. Open-source models like Kokoro are free to run on your own computer.
Usually, yes. Spotify asks you to tick a digital voice box and adds a disclosure, Apple labels its titles "Narrated by Apple Books", and Spotify's own tool labels them "Narrated by Digital Voice".
About 1,000 credits per minute of audio, so a 10-hour book takes roughly 600,000 credits, the monthly allowance on the $99 Pro plan.
