Preparing AI Audio for Distribution Platforms (Specs, Not Compliance)
Updated September 3, 2026
TL;DR: the common retailer spec: RMS between −23 and −18 dBFS, peaks no higher than −3 dB, 44.1 kHz, 192 kbps CBR MP3, one file per chapter under 120 minutes, quiet room tone at head and tail. LeapFun's Audiobook preset masters toward it and measures the encoded file, then shows a report. Meeting the spec is a technical fact; being accepted is a policy decision — ACX currently restricts third-party AI narration, and other retailers set their own rules.
The spec, item by item
| Item | Common requirement | What LeapFun does |
|---|---|---|
| RMS level | −23 to −18 dBFS | Audiobook preset masters to a −20 LUFS starting point; the encoded file is measured and reported |
| Peak | ≤ −3 dB | Preset ceiling −3.5 dBTP before encoding (MP3 raises peaks); measured after |
| Sample rate | 44.1 kHz | Preset outputs 44.1 kHz (Standard preset is 48 kHz for video) |
| Bitrate | ≥ 192 kbps, constant | MP3 export at 192 kbps CBR; the report checks the frame header, not just the average |
| File length | ≤ 120 minutes per file | Export per chapter; the report flags long files |
| Noise floor | ≤ −60 dBFS | Reported as an estimate only — silence between paragraphs is digital zero, which fools simple measurements |
| Head/tail room tone | 0.5–1 s head, 1–5 s tail | Add in your editor or via chapter padding; retailer specifics vary |
Reading the report
Each item shows the measured value, the target and a status: pass, fail or review. "Review" means the value couldn't be measured reliably (noise floor always shows this) — it isn't a hidden fail. The report also records which mastering profile ran and whether it fell back, so a file that skipped normalisation says so.
What the report does not say
It never says "compliant" or "certified", because a technical pass isn't a submission guarantee. Retailer policy on AI narration is separate: as of this writing, Audible's ACX does not accept third-party AI narration; other retailers and aggregators have their own positions and change them. Read the current policy of each platform before you produce a book for it.
Preparing files
- Export per chapter with the Audiobook presetOne file per chapter, MP3, named as the retailer lists chapters.
- Read the report and fix what failsPeak or RMS out of range usually means an unusual voice level; regenerate the loud paragraph or use a calmer director's note, then re-export.
- Add room tone and metadataHead and tail silence, title and author tags, cover art — in your editor or the retailer's uploader.
- Check policy, then uploadTechnical pass plus a platform that accepts AI narration for your title.
Streaming is different
Podcast hosts, YouTube and short-form platforms want −14 LUFS at 48 kHz — the Standard preset — not the audiobook spec. Don't send a −20 LUFS audiobook master to YouTube; it will be quiet.
Frequently asked questions
Does passing the checks mean I can submit to Audible?
No. The checks are technical. ACX currently restricts third-party AI narration regardless of file quality; the report says so. Check the policy of each retailer you target.
Why is noise floor only an estimate?
Generated narration has digital silence between paragraphs, which makes simple 'quietest window' measurements read as perfect. A reliable figure needs voice-activity analysis, so the report shows an estimate and never a pass/fail.
Can I use the Audiobook preset for a podcast?
You could, but streaming platforms expect −14 LUFS; use the Standard preset for podcasts and video.
What if a retailer wants WAV?
Export WAV with the Audiobook preset; the sample rate and peak handling are the same, and the report measures the WAV.