Booth Log entry
Passed the meter, still returned
It's the most demoralizing email in home narration: chapters that pass every measured spec — RMS in the window, peaks under −3, floor comfortably below −60 — flagged for audio quality anyway. It isn't a glitch and it isn't bad luck. It's the difference between the two reviews your files face: one performed by meters, one by ears, and only the first one publishes its numbers.
The spec has a listened half, in writing
Read ACX's requirements past the famous numbers and the listened half is right there (our verification date: 2026-09-20): submissions must be free of extra sounds — the spec names plosives, microphone pops, mouse clicks, excessive mouth noise, and outtakes — and must hold consistent sound across the book: levels, tone, noise, spacing, pronunciation. Not one of those clauses has a threshold a plugin can certify. That's not sloppy spec-writing; it's an honest admission that audiobook quality is ultimately a listening experience, and the meters exist to guard its lower bound. The numbers are the floor of quality. They were never the ceiling.
The measured half
RMS in the window, peaks under −3, floor below −60
Meters — thresholds a plugin can certify
The listened half
Free of extra sounds: plosives, pops, mouse clicks, mouth noise, outtakes. Consistent levels, tone, noise, spacing, pronunciation.
Ears — no threshold to certify
What ears catch that averages forgive
The pattern repeats across narrator forums and post houses, and it comes down to arithmetic: RMS measurements are averages over time, so anything brief barely dents them. A mouth click rides a syllable for 30 milliseconds — statistically invisible, perceptually a spotlight, and the spec names it explicitly. A single un-deleted outtake buried at a chapter's end moves no needle at all; QA plays files to their ends. Edit seams where room tone jump-cuts — tone patched with digital silence, or a floor that steps up mid-chapter because the fridge came on between sessions — measure fine within any window while announcing themselves at every boundary. And the over-processed floor is its own genre of return: a chain cranked until the meter reads spotless leaves pauses that don't breathe and consonants with that unmistakable underwater edge. The meter says immaculate; the ear says something happened to this recording. Cranking software until the number passes is how files start failing the review that matters — the reasoning lives in the chain guide's “less than the slider allows” rule.
The review you can run first
You can't run ACX's ears, but you can run yours, and a disciplined pass catches nearly everything the meter missed. The routine we'd stake a submission on:
- Meters first, so ears do only ear-work: floor, RMS, peaks on every file. Green numbers or back to the chain.
- Pauses at full volume. Headphones, volume genuinely up, listen between phrases at the head, middle, and tail of each file: tone should be continuous, quiet, and alive — no steps, no dead patches, no flutter.
- One full file at reading speed, doing nothing else. Not skimming — the outtake you left in is precisely where attention wasn't. Rotate which chapter gets the full-attention pass per session so the whole book eventually gets covered ears.
- Boundaries and seams: first and last ten seconds of every file (where room tone rules are checked and edits cluster), plus any spot you remember patching.
Total cost for a typical chapter batch: an evening. Total cost of a return: the same evening, plus re-mastering, plus the schedule slip, plus the confidence dent. The listen is cheaper.
The moral, because Log entries get one
Meters are how you negotiate with a spec; ears are how you negotiate with listeners. The measured half of ACX's requirements tells you your room and chain are under control — the fix order gets you there. The listened half asks a different question: would a stranger enjoy ten hours of this? Pass both reviews at your desk, in that order, and the demoralizing email stops arriving.