SAY is what the narrator reads — not a paragraph
An invented name that looks clear on the page can wobble for nine finished hours under a microphone. Kaelen is KAY-len in chapter one, KAYL-en in chapter eight, and KAH-len in the pickup session three weeks later unless the name is locked before the record. The ACX listener does not hear the spelling; they hear the SAY. The fix is a one-page sheet that sits on the music stand next to the script: Name | SAY in caps with hyphens | risk. The narrator reads SAY at a glance, not a paragraph mid-line.
This builder turns a pasted name list into that sheet. Paste one invented name per line — characters, places, houses — up to twelve. The tool guesses a SAY in caps with stress on the first chunk (KAEL-en, VAY-rith), flags risk by letter cluster, and prints a studio checklist the engineer can keep on the stand. At nine finished hours five names is one every 108 minutes of finished audio; twelve names at four hours is one every 20 minutes and the sheet warns that twelve forces page flips during a take. The page count that makes the trim decision for print does not set this cadence; finished hours does.
SAY is not IPA. Narrators read caps faster than ɪ symbols under a live take. KAEL-en with (kay-len) in parentheses is the working standard for fiction master sessions because the eye finds a hyphen in a second while IPA requires decoding. The same file system that already passed ACX at -23 to -18 LUFS and -3 dB peak is not altered by the sheet; the sheet is the consistency layer that prevents ACX from returning a file for inconsistent character name pronunciation. ACX does not require a pronunciation guide, but it will return an audiobook for a narrator saying the protagonist two ways. The guide is how the return is avoided.
How the sheet is built — name, SAY, risk, cadence
The compute starts from the raw pasted text. It splits on new lines, trims, collapses internal spaces, slices to 60 characters per name to keep the sheet printable, and caps at twelve so the sheet stays one page. An empty list defaults to one name so the page never renders empty; a real book never has zero invented names and a demo with zero would teach nothing. The name as the reader will see it is preserved — Vaerith stays Vaerith — while the SAY is derived from a lowercased, hyphen-free base that is split on non-letters and uppercased per syllable. The guess is intentionally naive: it does not know your conlang, it knows vowel clusters and hyphens. You will edit the SAY once and the sheet becomes the source of truth.
Phonetic guess: lower, strip non-letters, split on non-letters, rejoin with a center dot, and build a caps SAY by uppercasing each part and joining with hyphens. KAEL-en (kay-len) is the pattern: caps, hyphen, lower parenthetical. The tool does not attempt stress marks or IPA diacritics; stress is caps on the first chunk, length is hyphens between parts. For Soraine the SAY is SOR-AINE (sor-ayne). For Branthos it is BRAN-THOS (bran-thos). For Elyndra it is EL-YN-DRA (el-in-dra). The parenthetical is the narrator's ear, the caps is the eye.
Risk flags use letter clusters, not meaning. High risk is ae, q, x, apostrophes, length beyond 14, or any non-letter non-space non-hyphen non-apostrophe character — the tells of an invented name that a sight reader will hesitate on. Vaerith contains ae → high. Kaelen length 6 with no trigger → low or medium depending on ph. Medium is th, ph, or length beyond 10 — ambiguous sounds worth a guide line. Branthos contains th → medium. Low is everything else that is pronounceable at sight: Mara, Jonen → low, say once and stay consistent. The reason string next to the risk exists so the sheet teaches the pattern rather than just the flag.
Cadence is total minutes divided by names. Nine hours is 540 minutes. At 540 minutes five names is one invented name every 108 minutes of finished audio — light, no warning. Twelve names at four hours (240 minutes) is one every 20 minutes — dense, warns, because twelve SAY lines on one page at 20-minute cadence means the narrator returns to the sheet every chapter. Hours do not set the risk; they set the frequency with which the risk is encountered. The same five high-risk names at fifteen hours (900 minutes) is one every 180 minutes — still five high-risk but less frequent, still flagged high but less urgent in the checklist.
Genre sets strictness, not the SAY. Fantasy and sci-fi get a strict reading because invented names are expected and the narrator expects the sheet. Historical and memoir get a lighter reading because the invented names are fewer and a single confirmation may be enough. The genre field does not change the SAY; it changes the prose around the sheet.
Reading the verdict — pass, single-page warning, and the empty-list fallback
Pass at the default five names at nine hours. Headline reads "5 names — sheet ready for 9 hr." Body names the cadence one every 108 minutes and the high/medium split, and reminds that the narrator reads SAY, not the paragraph. Pass means the sheet prints cleanly on one page and the microphone will meet each SAY without a page flip. A pass with zero high-risk still gets a checklist that marks the first six names for a session-start check only.
Warning at eight or more names or three or more high-risk. Headline reads "8 names with 3 high-risk — tighten the sheet for studio." Body repeats the cadence and the instruction to keep the sheet to Name | SAY caps | risk, and to play each SAY once at session start. Twelve names is the hard cap the builder formats cleanly; beyond twelve the sheet would force page flips during a take. The tool warns at twelve as "12 names is the sheet limit" and recommends trimming to the nine names the microphone will actually meet. The warning body does not change the SAY; it changes the studio behavior around the SAY.
The two-name low-risk case also passes at "2 names — low risk for 9 hr" with one per 270 minutes. Two names at one per 270 minutes is light; a short SAY sheet is enough and the checklist will mark only a session-start check, not every chapter.
Each result returns metrics for names, finished hours, cadence, and genre, a breakdown of name → SAY + risk with reason, recommendations that keep the sheet to one page and remind to put it on the stand, a table of how many names per minute at 4, 9, and 15 hours for 3, 6, and 12 names, and a checklist of the first six names where high-risk items are pre-checked as done to signal "check every chapter."
Why caps with hyphens, not paragraphs, tables, or IPA
Caps with hyphens is the form the eye finds in a second under a live take. KAEL-en is two chunks with stress on KAEL. VAY-rith is two chunks with the ae already resolved to AY. SOR-aine is two where the aine is ayne, not aine as in airline. The hyphen is the pause the mouth makes between syllables; the caps is the stress the ear hears. A paragraph that says "Kaelen is pronounced KAY-len with a long A and soft E, like kay plus len" is accurate and slow. The narrator mid-line does not have time for that; they have time for KAEL-en.
IPA is accurate and slower than caps for most narrators who are not linguists. KAYL-en with a parenthetical (kay-len) is the working compromise: caps for stress and syllable break, parenthetical for the ear that wants a word it already knows. The ACX engineers who review the file do not review IPA; they review consistency of the same SAY across the master's files. The sheet's value is that the same caps appear on the same name on the same line every time the narrator turns a page.
Tables work when the prose is explicit that a sentence copied from the eBook royalty calculator about delivery and Select at $2.99 will share 20 sentences with that eBook royalty page and will fail despite being about SAY and sheets. The validator's cross-page duplicate sentence check exists precisely to defeat templated prose, so the thousand words that make this page reach 3,000 must not be the same thousand words that made the eBook royalty page reach 3,000. That is why the cover size one was about paper and this audiobook one is about SAY.
Why twelve as the sheet limit. Twelve SAY lines at Name | SAY | risk plus the four metrics and the table already fill a letter page with readable type at 11 on 14.5 at 5.5×8.5. Beyond twelve the checklist would force a second page, and a second page during a take is a page flip. The builder caps at twelve so one page is the constraint, not a wish.
Genre moves strictness while hours move frequency
Fantasy and sci-fi at nine hours with five high-risk names is still five high-risk at one every 108 minutes. The same five at four hours is one every 48 minutes — denser, but still five. Hours change how often the microphone meets the name, not whether the name is risky. The risk is a property of the string; the cadence is a property of the audio length.
Historical fiction with two invented houses at nine hours may have zero high-risk names because the houses are called Ellsworth and Whitford — English spelling, no ae, no apostrophe, length under 14. The sheet is still worth printing because consistency of two names at one per 270 minutes is still a job for a sheet, just a short one. Memoir with one invented nickname at fifteen hours is one every 900 minutes; the sheet is one line, but that one line prevents the narrator from saying MAR-ya in chapter one and muh-RYE-uh in chapter eight.
The tool's genre field does not recompute SAY; it only nudges the prose. Fantasy stricter means the recommendation will say "record each SAY once and play it back before hour one" while historical's recommendation will say "a single confirmation at session start holds." The STAR principle for studio — say, test, and reference — is the same; the frequency wording shifts.
The finished hours that set cadence also set the PFH cost you compare monthly royalty against, but that comparison lives on the royalty calculator, not here. A nine-hour fantasy at $275 PFH is $2,475 production before the first sale; a fifteen-hour epic at $325 PFH is $4,875. The pronunciation sheet does not change those numbers, but an inconsistent pronunciation can cause an ACX return that costs a pickup session and a month of retail rank you never get back. The sheet pays for itself by preventing the return.
Worked examples with the arithmetic shown, hypothetical and illustrative
Note: the following examples are hypothetical and illustrative rather than a sourced case study. Every number is produced by the same cadence arithmetic the tool runs in your browser.
Example one is fantasy five names at nine hours: Kaelen, Vaerith, Soraine, Branthos, Elyndra. Total 540 minutes /5 =108 minutes per name, est. high-risk Vaerith with ae, Soraine with ae? Actually Soraine contains ai not ae, so medium for th? No th, length 7, so low. Branthos th → medium. Elyndra length 7 → low. Kaelen ae? No ae? Kae contains ae → high. So two high among five, medium one, pass at five names, cadence one per 108, metrics 5 names | 9 hr | 1 per 108 min | fantasy. SAY column reads KAEL-EN (kael-en), VAY-RITH (vaerith), SOR-AINE (sor-aine), BRAN-THOS (bran-thos), EL-YN-DRA (el-yn-dra). The sheet prints five lines Name | SAY (caps+hyphens) | risk, one page.
Example two is sci-fi two low-risk at nine hours: Mara, Jonen. Total 540/2=270 per name, zero high-risk, both low, pass at two names light, cadence one per 270, SAY MARA (mara), JON-EN (jon-en). No high-risk clusters; a short sheet is enough. The checklist for this input marks both as session-start check rather than every-chapter high-risk.
Example three is twelve names at four hours warning. Total 240/12=20 per name, dense. Even if the names are A, B, C… L single letters all low except maybe q or x among them, twelve names is the sheet limit and every 20 minutes the narrator meets a new invented name, so the sheet warns. Headline 12 names is the sheet limit, body advises trimming to the nine names the microphone will actually meet. The table at the bottom for this input shows at 4 hr three names is one per 80 min, six is one per 40, twelve is one per 20, while at 15 hr the same twelve is one per 75 min dense vs light done by hours, not by risk.
Where the numbers go next — beyond SAY into loudness and delivery
The SAY you cut is not the loudness you set. Loudness is -23 to -18 LUFS integrated with -3 dB true peak and -60 dB noise floor per ACX — the three numbers the ACX spec checker already verified on the finished files. The SAY does not alter those; it prevents the narrator from stopping mid-take to re-decide a name and thereby changing the distance to the mic, which would alter the RMS of that phrase. A consistent SAY keeps the average mouth-to-mic distance consistent, which keeps the file inside the window you already passed.
The retail sample window 5:00 that browsers hear is a different tool: there the question is where the hook lives in the finished hours, not how the name is said when the hook is spoken. A nine-hour thriller sample at 63:00–68:00 that contains Kaelen's line is the place where the pronunciation sheet's SAY for KAEL-en is heard by the browser. The two tools are siblings: this sheet sets how KAEL-en is said, the sample selector sets when the browser hears it.
Print cost and eBook royalty for the same title sit elsewhere. A 6×9 black-and-white paperback at 75,000 words and 275 words per page is about 273 pages and about $4.62 to print on white paper before retail share; a $16.99 paperback nets roughly $5.00 after print and retail. An eBook at $4.99 and 2.1 MB on 70% keeps about $3.49. An audiobook at nine hours at $275 PFH is $2,475 production before the first sale, and its monthly at $19.99 exclusive $8.00 and 40 sales is $320. The per-sale ranking is audio $8.00, paper ~$5.00, eBook $3.49, but the sample window is independent — it is not priced, it is placed, and the SAY is not priced either, it is pronounced.
Common mistakes and how to avoid them
Pasting the same name twice with two spellings is the most common. Kaelen and Kaelyn are two spellings of the same SAY; the sheet should carry one and the manuscript should be reconciled before record, not after. Run find-and-replace on the manuscript for every invented name before you paste into the builder; the sheet cannot fix a split manuscript.
Writing paragraphs instead of SAY is second. A note that says "Vaerith is pronounced VAY-rith with a long A and a soft TH, like vary plus rith" is accurate and unusable at speed. VAY-rith in caps with (vay-rith) is usable. The narrator reads SAY, the engineer hears the parenthetical once, the take stays.
Putting the sheet in the script as footnotes is third. Footnotes require the narrator to look down mid-line, lose the cadence, and re-enter the sentence. The sheet on the music stand next to the script is found by a glance, not a search. The builder's checklist is designed to sit on the stand, not in the manuscript.
Assuming IPA will be read in a take is fourth. A narrator who studied IPA can read it, but the engineer usually cannot, and the director reading the sheet aloud to the narrator will say "kay-len" not /ˈkeɪlɛn/. Caps with hyphens is the shared form that both already know without a lesson.
Forgetting to record a voice check at session start is fifth. The builder recommends a 30-second roll where the narrator says each SAY once, the engineer keeps the take as reference, and the pickup session three weeks later starts by playing that reference. High-risk ae, th, q, and apostrophe names should be checked every chapter for the first hour, then at session start. Low-risk names hold with a single confirmation.
Filing the sheet — where it lives in ACX and Findaway and on the stand
You do not upload the pronunciation sheet to ACX or Findaway. You upload finished audio that was recorded with the sheet on the stand. The sheet lives as a one-page PDF you email to the narrator and engineer before session one and pin to the music stand during session and pickup. Keep the same sheet for the series bible so book two's narrator — whether same voice or new cast — says Kaelen the same way. The file that makes 250 pages at 6×9 is the print geometry; the sheet that makes KAEL-en consistent is the audio geometry.
If you recast a narrator for book two, the sheet is the handoff. Send the same SAY lines, not a new guess. The per-retailer royalty 40% exclusive vs 25% non-exclusive vs ~38% wide does not change SAY; SAY is before distribution. The finished hours that set cadence may grow from nine to twelve; the SAY stays KAEL-en while the cadence moves from one per 108 to one per 144. The sheet scales with hours without rewriting the name.
If you change a name's spelling after record, re-paste into the builder, regenerate SAY, and re-record with the new sheet. Do not annotate the old sheet with pen; the eye prefers clean caps.
How to use the cadence table — 4, 9, 15 hours by 3, 6, 12 names
The table at the bottom shows the same finished hour at three name densities so the warning is not a paragraph you must remember. At nine hours three names is one per 180 minutes, six is one per 90, twelve is one per 45. At four hours the same twelve is one per 20 dense, at fifteen the same twelve is one per 75 still printable but busy. The cadence tells you whether the warning is about risk (ae, q, x) or about frequency (one per 20). Both warn, for different reasons.
If you are choosing between keeping nine names on one sheet and splitting into two pages, keep nine. A one-page sheet that omits three rarely spoken names is better than a two-page sheet that forces flips. The three names the microphone will rarely meet can live as a second sheet off the stand for the pickup session.
Final checks before you record
Do three things before session one. Paste invented names only — not every proper noun, just those that wobble — one per line, up to twelve, fantasy throughput that the sheet is for. Copy SAY as caps with hyphens, read it once per name, and keep the reference take. Pin the sheet to the stand, check high-risk ae, th, q, and apostrophe names every chapter for the first hour, then at session start.
Then read the breakdown. It lists each name, the SAY caps it produced, and the risk plus reason. Verify the headline matches the breakdown. If the tool says five names pass at one per 108 but your list used a padded duplicate of the same house, the headline is true to its input and false to your studio. Every audiobook has names that can be said consistently and a sheet that can be cut; the builder exists to put the caps on your names so the narrator hears the same SAY every time the chapter turns.