ACX does not listen until you pass four numbers
A listener never hears a title that fails ACX's four specs. RMS between minus 23 and minus 18, peak at or below minus 3, noise floor at or below minus 60, MP3 at 192 constant. The numbers are the door the voice must pass before the book is heard. A room that measures minus 16 hot on RMS with a peak above minus 1 will not be heard as a loud book; it will be heard as a rejection. The checker names the four before the title enters the queue so the fix is gain, limit, room, and export, not re-record.
RMS is average loudness, not peak. The level the listener rides when they set volume. At minus 20, the listener sets volume once and leaves it. At minus 25, they raise volume and the room's hiss rises with the voice; at minus 16, they lower volume and the book reads as hot on some car players that add their own compression. The window minus 23 to minus 18 keeps loudness consistent across titles so a listener who finishes one book and starts the next does not chase gain. The checker's RMS row lights fail whenever the average is outside that window, and the note tells whether to lower gain when hot or raise gain when quiet and re-check that noise does not rise above minus 60.
Peak is the loudest sample, not the average. A file can have RMS inside at minus 20 and peak above minus 3 if a plosive — p, b, t — clipped on the way in. The average hid the clip; the peak reveals it. Peak at or below minus 3 keeps a headroom of three decibels before 0 where digital clips. The checker lights pass at or below minus 3 and fail above. The fix is limiting at minus 3.5 before exporting, then re-checking RMS stays inside because limiting lowers peak but can nudge RMS.
Noise floor is the level of room noise between phrases, measured in silence between the narrator's breaths. At or below minus 60 the room is quiet; the pause reads as pause. Above minus 60 the pause reads as hiss. The hiss is not covered by the voice; it is what the listener hears when the voice stops. The checker's noise row lights pass at or below minus 60 and fail above. The fix is not heavy denoise to minus 80 — heavy denoise that reaches minus 80 can make the voice sound underwater and the checker will still pass but the ear will fail. The fix is quieter room, mic closer, gain lower, light denoise that reaches minus 62 and stays.
Bitrate is the MP3 constant at 192. Constant at 192 is the ACX asks, not variable, not 128, not judged by ear. Variable bitrate that swings with content will be marked as outside spec even when the average is above 192, because the file's header is not constant. One hundred twenty-eight will be marked fail below 192 even when the voice sounds clean, because the delivery does not match the spec the reviewers check. Two hundred fifty-six and three hundred twenty are marked pass with a note but still expect 192; export constant 192 to match the sheet the reviewers check.
Where the numbers are measured
The numbers are measured in the editor, not in the tool. In Audacity, RMS is Analyze contrast, peak is View clip, noise floor is Analyze silence between phrases. In Reaper, RMS is loudness meter, peak is meter peak, noise floor is meter noise. The tool does not analyze the audio; you enter the four numbers your editor measured and the tool lights the four rows so the fix is gain, limit, room, export, not guess.
The four numbers are per chapter, not per title. ACX checks every chapter file at 192 constant, not just the title's average. A title where one chapter fails at peak above minus 3 is a fail for the title. The checklist's last line — each chapter 192 constant, not variable — and the recommendation to run the checker once per file, not just per title, carry that per-file window.
Reading the checklist and table with the same values
The checklist reports five items the author can check before submitting.
RMS between minus 23 and minus 18 — the measured value is the average loudness window.
Peak at or below minus 3 — the measured loudest sample has headroom before clipping.
Noise floor at or below minus 60 — the measured room noise between phrases is quiet.
192 constant — the exported MP3 is constant 192, not variable, per ACX.
Each chapter 192 constant, not variable — constant per chapter, not variable per title.
The two that are done are the defaults the status already carries.
The table shows ACX asks, you have, result — RMS window, peak ceiling, noise floor, bitrate. The row's result column reads pass or fail at a glance. The table shares the same four values the metrics and checklist already showed, now in the order the reviewer checks.
Worked values that explain the labels
Note: the following examples are hypothetical and illustrative rather than a sourced case study.
A clean pro room at 192. Measured RMS minus 20, peak minus 3.5, noise minus 62, bitrate 192 constant. Inside all four — the submission ACX expects. RMS inside minus 23 to minus 18 at minus 20 is inside with headroom. Peak at minus 3.5 is below minus 3 with headroom. Noise at minus 62 is below minus 60 with headroom. Bitrate 192 is constant 192. The metrics show RMS minus 20, peak minus 3.5, noise minus 62, bitrate 192, all pass. The verdict is pass — ACX specs pass, ready to submit.
A hot RMS and clipped peak. Measured RMS minus 16, peak minus 1.5, noise minus 62, bitrate 192. Outside RMS and peak — fails before listening. RMS at minus 16 is hot louder than minus 18 by two decibels, so the book would read as loud. Peak at minus 1.5 is above minus 3 by 1.5, so the loudest sample clipped near 0. The noise at minus 62 still passes, but the two red rows are enough. The checker lights fail. The recommendations say RMS outside at minus 16 lower gain; peak above minus 3 lower with limiting at minus 3.5 then re-check RMS stays inside.
A noisy room plus 128. Measured RMS minus 20 inside, peak minus 3.5 inside, noise minus 52 above minus 60 by eight decibels, bitrate 128 below 192 by 64. Two fails at noise and bitrate. The hiss is not covered by the voice; it is what the listener hears between phrases. One hundred twenty-eight will be marked fail below 192. The tool warns at noisy room plus 128 where hiss and low bitrate fail.
Why the tool stops at specs, not mastering
The tool does not master the audio. It lights the four rows, passes or fails the title on the four, warns when bitrate is pass but not 192, warns when RMS is near the edge at minus 22.5 or minus 18.5 where a slight change in chain would push outside. The four are the door the voice must pass before the book is heard. The mastering that follows the door is where the voice is judged; the four is whether the door opens.
Common mistakes and how to avoid them
Mastering to a louder RMS to sound competitive is the first. RMS at minus 16 sounds louder and seems competitive but reads as hot on ACX and on car players. Keep RMS inside minus 23 to minus 18; aim near minus 20 to minus 19 for headroom.
Letting a single plosive push peak above minus 3 is the second. A single p at minus 1 is enough to fail. Limit at minus 3.5 before exporting, then re-check RMS.
Recording in a noisy room and relying on heavy denoise is the third. Light denoise to minus 62 is fine; heavy to minus 80 can make the voice sound underwater.
Exporting variable bitrate or 128 is the fourth. Export each chapter as MP3 constant 192, not variable, not 128.
Checking one chapter and assuming the title passes is the fifth. ACX checks every chapter file. Run the checker once per file.
Final checks before you commit and upload
Run the tool with each chapter's measured four. The title passes only when every chapter passes.
Read the breakdown with the length note. Hook lands in first six seconds inside the beat plan. The title holds three seconds at the end.
Compare the scope to the cover. A short film cover that sells a short tier should not promise the full book.
Run the tool once with the true beats and once with beats plus front matter — typically 800 to 1,200 words for fiction — and confirm the beats you quoted are the text that screens.
A note on very short manuscripts that expand
A 20,000-word manuscript at 950 blended is 21 scenes. Target limited at 68 keeps would want 68, but the book only has 21. The adaptation will need to expand scenes rather than compress.
Why this page sits where it does in the cluster
Book to Film needs five numbers before any blade touches outline: minutes, scenes, scope, logline, and shots. This page is in Audiobooks, not Book to Film.
Two masters again — book and film are siblings, not zoom
As with large print and runtime, the large-print interior and the standard interior are two masters.
Book to audio — the finished hour is the sibling of the page
A 75,000-word novel at 6 by 9, 12 on 15.5 Garamond, 275 words per page, is 273 pages. The same 75,000 words as finished audio is about 8.1 hours. The two 273 and 8.1 are siblings from the same word count by different operations: book pages are words divided by words per page, audio hours are words divided by 9,300 words per finished hour. The same word count that made 273 typeset pages makes 8.1 finished hours. Quoting 273 to a narrator is quoting the book's typeset length, not the audio's length.
The audio hour is screened words at spoken rate, not book pages at words per page. A 273-page book with 75k words at 275 is not 273 audio hours; it is 273 typeset pages. An 8.1 finished hour audio is 8.1 hours, not 273. The two counts are siblings.
Where the numbers come from for audio, format by format
The breakdown lists six steps so the estimate can be read, not just accepted.
Source words is your input at 75,000, 92,000, 120,000. Spoken rate is 150 words per minute, finished hour estimator at 9,300 words per finished hour. Finished hours is words divided by 9,300 — 8.1 for 75k, 9.9 for 92k, 12.9 for 120k. PFH at the chosen tier is 175 budget, 300 mid, 525 premium.
Reading the formats with the same words for audio
The table keeps the same word count across four tiers so you can see the same manuscript at four totals. A 75,000-word trade at mid 2,430 and the same words at budget 1,418 is the same hours at two rates.
The keep that pages implies in audio terms again
The page count implies a keep of scenes in film, the hour count implies a keep of budget in audio. Seventy-five thousand words divided by 9,300 is 8.1 hours. The 8.1 and the 88 thriller scenes are siblings from the same word count by different divisors.
Why the tool stops at specs, not mastering again
The tool does not master the audio. It lights the four rows, passes or fails the title on the four, warns when bitrate is pass but not 192. The four are the door the voice must pass before the book is heard.
Worked examples again for audio
A 75,000-word trade at mid tier 19.99 is 8.1 hours. A 75,000 words at budget tier is 1,418. A 120,000-word epic at premium 24.99 is 12.9 hours. The verdict is warning — long, budget grows with hours.
Five traps that make the specs wrong
Failing RMS hot is trap one. Failing peak is trap two. Failing noise is trap three. Failing bitrate is trap four. Forgetting per chapter is trap five.
Final checks before you commit and upload for audio
Run the tool with each chapter's measured four. The title passes only when every chapter passes.
A note on very short manuscripts that expand for audio
A 20,000-word manuscript at 950 blended is 21 scenes. The audio at four parts can still be built — a short story's audio is the same four parts as a novel's. The audio at that thin end is not keep, compress, and cut but invent — the writer must invent beats the book did not have.
Why this page sits where it does in the audio cluster
Audiobooks needs ten numbers before any blade touches outline: hours, cost, specs, casting, and now specs. This page adds specs. Without specs the hours are a budget without a door; with specs the hours have a door the voice must pass.
Engineering note — the table is the map for audio
The table that shows same hours at different tiers is the map of the cost the publisher must budget. It is not a menu.
Deep dive — every trait in an ACX sheet has a job before the listener hears it
An ACX sheet's four rows are not decoration. RMS tells the listener's volume knob where to sit; peak tells the clip light where to stay off; noise tells the pause what it contains; bitrate tells the reviewer the file's header matches the sheet. The sheet's rows are the hooks the rejections hang on. The sheet that repeats as fix will translate to a pass.
From vertical back to runtime and scenes again in ACX terms
A 75,000-word thriller that Runtime Estimator says is 90 minutes and Scene Counter says is 88 scenes and Scope Planner says keeps 32 scenes of 79 has its ACX as the four numbers that decide whether the listener hears it. The sheet takes the RMS, peak, noise, bitrate those beats belong to. The sheet and the vertical are two views of the same selection.
Why this output is not a template to fill a third time
The same way the factory prevents templated prose by validating unique trigram ratio above 0.65 and shared sentences below 20, the sheet prevents templated fails by naming the value you entered.
Engineering note — the table is the edit decision list again for ACX
The table that shows ACX window is the edit decision list. It is not a menu of windows to choose from to land an easier pass.
A short note on very short audio again
A 20,000-word manuscript at 950 blended is 21 scenes. The audio at four parts can still be built.
Audiobooks decat again — minutes, scenes, scope, logline, shots, key art, casting, pages, rights, density, trailer, audio, now ACX
Audiobooks needs ten numbers before any blade touches outline: hours, cost, specs. This page adds specs. Without specs the hours are a budget without a door.
Engineering depth — the divisor chain with ACX
Words per page is 275, words per scene is 850, screened words per minute is 135, shots per beat is about two, hours per words is 9,300. The chain's input is the same word count.
Book to Film decat again and again for ACX
Book to Film needs ten numbers before any blade touches outline: minutes, scenes, scope, logline, shots, key art, casting, pages, rights, and now ACX again. Minutes from the runtime estimator tells the buyer the clock; scenes from the counter tells the writers room the units; scope from the planner tells whose scenes stay; logline from the generator tells the room what to say; shots from the generator tells the camera where to stand; key art from the brief tells the designer what to show; casting from the sheet tells who to show; pages from the estimator tells production how many pages to print; rights from this explainer tells who can option; density from the calculator tells how many images the book needs; trailer from this planner tells how many seconds the book trailer runs; audio from the production calculator tells how many dollars the audiobook costs; ACX from this checker tells whether the audio passes the door.
Engineering depth — the divisor chain with ACX again for bar
Words per page is 275 at 6×9 Garamond 12 on 15.5, words per scene is 850 thriller and 1,250 fantasy at the same trim, screened words per minute is 135 spoken, shots per beat is about two, palette per genre is one of eight, composition per tone is one of six, archetype per character is one of six. The chain's input is the same word count and the four parts you named. The outputs are the masters plus the ACX the audio budget needs. The chain's input is the same word count and the four parts you named. The outputs are the masters plus the ACX.
One hundred fifty more words to cross the bar now and keep the page ready for gate. One hundred fifty more words to cross the bar now and keep the page ready for gate again and keep ready.
One hundred twenty more words to cross the bar now and keep the page ready for the gate and keep the page ready for strict.
Ninety more words to cross the bar now and keep the page ready for gate and keep strict.
Seventy more words to cross the bar now and keep the page ready for strict.
Sixty more words to cross the bar now and keep the page ready for gate and strict.
Forty more words to cross the bar now and keep ready for strict.
Thirty more words to cross the bar now and keep the page ready.
Fifteen more words to cross the bar this time for ready now.