Reference · every platform's own words
Audiobook audio requirements by platform
ACX is no longer the only door. Spotify sells audiobooks directly, Google Play takes uploads, Apple routes through partners — and each publishes a different amount of specification. This table carries only what each platform states itself, links the source, and says “not published” where that is the truthful cell. Audio rules last verified 2026-08-31; the AI-narration column read 2026-09-23.
The table
What each platform publishes, and where it stops
| Platform | Formats | Levels | Rate / bitrate | Structure | AI narration made elsewhere |
|---|---|---|---|---|---|
| ACX / Audible | MP3, 192 kbps or higher, constant bitrate | RMS −23 to −18 dB · peak < −3 dB · noise floor < −60 dB RMS | 44.1 kHz | one chapter per file · ≤ 120 min · 1–5 s room tone both ends, and room-tone spacing must not exceed 5 s · all mono or all stereo | not allowed unless authorised — details |
| Voices by INaudio (formerly Findaway) |
constant bitrate, 192 kbps or higher | RMS −23 to −18 dB · peak below −3 dB · maximum −60 dB floor | 44.1 kHz | ≤ 120 min · 0.5–1 s silence at the head, 1–5 s at the tail — stated as a requirement, not advice | only unmodified packages from three named tools — details |
| Spotify for Authors | MP3 preferred (≥192 kbps, CBR preferred), WAV, FLAC | RMS −24 to −14 dB · no peak limit published · noise floor under −60 dB — Spotify files all of these under “Recommended qualities”, not requirements | 44.1 kHz, 16-bit — stated for MP3 and WAV; Spotify publishes no rate or depth for FLAC | ≤ 120 min per file · 0.5–1 s head, 1–5 s tail · mono or stereo, never mixed | only unmodified packages from three named tools — details |
| Author’s Republic (distributor) |
MP3, 192 kbps, constant bitrate | RMS −23 to −18 dB · peak no higher than −3 dB · floor no higher than −60 dB RMS | 44.1 kHz | ≤ 119 min · ≤ 170 MB per file · 1–5 s silence both ends · one channel format across files | not checked for this page |
| Storytel | .wav or .mp3, never mixed in one title; mono 128 kbps or stereo 256 kbps | −15 to −18 LUFS, peak no higher than −0.3, maximum noise floor −80 — their document writes all three in “LUFS”, which is a unit slip for the peak and floor | 44.1 kHz — written “44,100 kHz” in their document, a second unit slip | no duration or silence rules published · no hard size limit, but “moderately sized files” recommended · sample files must not be submitted | not checked for this page |
| Google Play Books | .mp3, .aac, .flac, .wav, .zip, .lpf | none published | MP3 ≥128 kbps mono / ≥256 kbps stereo, CBR preferred · FLAC and WAV 16-bit at ≥44.1 kHz | 5 min to 100 h for the whole audiobook · files named ID_XofY (Google’s recommended scheme) · one supplemental PDF under 100 MB | offers its own free auto-narration; outside AI files not confirmed — details |
| Apple Books | WAV (≤2 GB), ALAC, AAC, FLAC, MP3 — MP3 least preferred | none published | 22.05 kHz 16-bit minimum, or 96 kHz 24-bit · bit rate constant throughout | under 23 hours: one single track · 23 hours or more: at most 25 tracks, each at least 60 minutes · chapters must not span track boundaries | offers its own digital narration; silent on outside AI files — details |
| Kobo | MP3 only | none published | “most audio is compressed to 64kbps” on ingest, so Kobo says higher fidelity is “not necessary” | 200 MB per file, 2 GB for all files combined | accepted if labelled as a synthesised voice — details |
Sources, each the platform’s own page: ACX audio submission requirements · Spotify for Authors, uploading audiobooks and the metadata and assets guide it links · Voices by INaudio technical requirements · Author’s Republic audio requirements · Storytel audio file requirements · Apple Books Audiobooks Specification · Kobo Writing Life audiobook files · Google Play Books audiobook file guidelines · Apple Books for Authors.
Text-to-speech and voice clones
Can I publish AI narration here?
What each platform’s own page says about an audiobook narrated by an AI voice made with a tool of your choosing. Every page below was read 2026-09-23; where a platform says nothing, this says so rather than guessing. These rules change — read the linked page again before you decide.
ACX / Audible: no, unless authorised
ACX’s requirements say AI or text-to-speech narration is prohibited unless authorised. The authorised routes are Amazon’s own: KDP Virtual Voice, which is invite-only; Audible’s publisher programme; and the ACX Voice Replica beta, also invite-only, in which ACX builds the replica itself — so a clone you made elsewhere is still unauthorised. The same page adds that Audible is working to accept third-party TTS content, with no date given.
Sources: ACX audio submission requirements · KDP Virtual Voice · ACX Voice Replica beta · Audible’s publisher announcement.
Spotify and Voices by INaudio: only three named sources
Digital-voice narration is accepted only as unmodified LPF packages from Google Play Books, ElevenLabs or Spoken Press. Audio from any other AI tool, or a package you have edited, is outside that rule.
Sources: Spotify for Authors, digital voice narration · Voices by INaudio.
Google Play Books: its own auto-narration
Google offers free auto-narration of its own. Whether it accepts an audiobook narrated by an AI voice made elsewhere is not something we could confirm on a Google page, so this does not claim either answer.
Apple Books: its own digital narration
Apple offers digital narration of its own. Its page is silent on audiobooks narrated by AI voices from elsewhere, so the answer for those is: not stated.
Source: Apple Books digital narration.
Kobo: yes, if it is labelled
Kobo says it will gladly accept AI-narrated audiobooks, provided the narration is labelled as a synthesised voice.
Source: Kobo Writing Life.
What this means in practice
An audiobook narrated by an AI voice from a tool of your choosing can go to Kobo and to direct sales, but not to Audible, and not to Spotify unless it came from one of the three named sources. Google and Apple each offer their own narration instead. Every tool on this site measures a recording the same way whoever or whatever read it, and none of them makes one.
The practical question
Does an ACX master pass everywhere else?
Mono: yes, almost everywhere it is specified
A mono 192 kbps CBR MP3 at 44.1 kHz — what the chapter fixer exports — clears Google Play's ≥128 kbps mono floor with room to spare, and MP3 is on Spotify's accepted list. Where a platform publishes no numbers, an ACX‑compliant file is the conservative choice, not a guaranteed one.
Stereo: watch Google's higher floor
Google Play asks stereo MP3 for ≥256 kbps. A 192 kbps stereo file meets ACX and misses that floor — the one place the same bytes pass one platform and fail another on the same axis. Re‑export stereo at 256 for Google, or deliver mono.
Lossless originals travel best
Google and Spotify both take FLAC and WAV. If you still have the edited master, sending the lossless file lets the platform do its own encoding instead of re‑compressing an MP3. Keep the master; the MP3 is a delivery, not an archive.
Apple inverts the file-structure rules
Everyone else wants one chapter per file, up to two hours. Apple starts from the other end, and the rule turns on the length of the book. Under 23 hours, Apple says the audiobook must be delivered as a single track. Only at 23 hours or more may it be split, and then at most 25 tracks, each at least 60 minutes. So a ten-hour book split into 30 chapter files — a perfectly conformant ACX delivery — fails Apple not on chapter length but on being split at all; the 60-minute floor never engages at that length. The batch checker flags whichever of the two applies when you check a whole book.
Apple takes no direct upload either: its own page says to publish with the help of one of our preferred distribution partners, so in practice the partner’s specification is the one your files must satisfy first.
Three incompatible rulers
ACX, INaudio, Spotify and Author’s Republic specify unweighted RMS; Storytel specifies gated LUFS. They are different measurements of different things, and no standards body publishes a conversion between them — a file at −20 dB RMS has no single LUFS value. The chapter checker uses the RMS ruler, the loudness tool the LUFS one, and neither borrows the other’s thresholds.
Podcast targets are a different ruler
Podcast platforms measure integrated loudness in LUFS, not RMS — Apple's podcast audio requirements ask for −16 dB LKFS within a decibel with true peak under −1 dBTP. The loudness normalizer measures and corrects against that ruler; the ACX numbers on this page do not transfer to it.
Check before you upload
Measure the file you are about to send
The chapter checker reads the five ACX numbers from your file's own bytes and fixes the fixable ones; the batch checker runs the whole book and flags chapters that do not match it, including a mix of mono and stereo files. Both run in this tab — an unreleased manuscript under NDA never leaves your machine.