The audiobook mistake listeners forgive least is a mispronounced name, and it is the one authors find last. It sits in chapter one, sounds fine to a narrator who has never met the word, and gets repeated sixty times before anyone with a stake in it presses Play. By then it is threaded through every scene the character is in.
Until this week, the way to prevent that on Midsummerr was to know your risky words in advance and enter them on the Pronunciation page before generating. That still works. What changed is that you no longer have to do the knowing yourself.
What happens at upload now
When you upload an English-language manuscript, Midsummerr reads it for the words a narrator could plausibly say more than one way: character names, place names, surnames with an ambiguous vowel, invented terms, and the stylized spellings that fantasy, romance and historical fiction run on. Shortly after upload, before the first chapter is produced, they appear on the Pronunciation page as a suggested list, ordered by how often each word occurs in the book.
Each row shows the word, a suggested reading written both as a plain respelling and in IPA, and how many times it appears in the manuscript. The count is the priority: a surname that appears two hundred times is worth a listen; a place mentioned once in chapter thirty is not.
The list is a first pass, not a verdict. It exists so that the words most likely to go wrong are in front of you before any of them is spoken.
Ready to try it on your own book?
Start your first chapter free →Hear it as written first
Every suggested row starts on "as written". Press Play and you hear the word exactly as the narrator reads it today, in the narrator's own voice, inside a short sentence. If that is already right, and it often is, you leave the row alone and move on. Nothing changes.
If it is not right, cycle through the suggested readings. Each one plays in the same voice, so you are comparing like with like: not a respelling on a screen against a guess in your head, but takes of the same name in the voice that will carry it through the book. Pick the one that matches what you have always heard in your own head, press Use, and it becomes a confirmed entry.
Take Carmilla, Le Fanu's Styrian vampire novella, which we ran through the new step as a test. Thirteen words came back, led by Carmilla itself, Karnstein, Mircalla, Millarca, Spielsdorf and Styria. For the title character, the suggested readings were kar-MIL-uh, kar-MEE-luh and kar-MY-luh. None of those is wrong. Only the author knows which one is the book, and that is exactly the decision this step puts where it belongs.
Foreign names in their own sounds
Names that come from another language are the hardest case, because the reading an English-speaking listener expects and the reading a native speaker expects are different, and both are defensible. So when a flagged word is French, German, Spanish or otherwise not English, the first suggestion is written with that language's own sounds, and the readings behind it are English-friendly approximations.
You choose between them by ear, not by theory. A Spanish surname in a thriller set in Madrid may want the native reading; the same surname in a family saga set in Chicago may want the anglicized one the characters themselves would use. Both are one Play and one Use away.
And if none of the suggestions is what you want, the entry is yours to write. Edit pre-fills the modal with the reading you were looking at; change the respelling, or type IPA between slashes, and a typed IPA reading is used exactly as you wrote it. Generate a test take, confirm it, and it joins the list like any other entry.
Nothing reaches the narrator unreviewed
A suggestion is not a setting. Until you press Use, it has no effect on production: no chapter is generated with it, no take is rendered from it. Dismiss a row and the word stays as written. Ignore the whole list and your book is produced exactly as it would have been last week.
When you do confirm a reading, it works the way the Pronunciation page has always worked. The entry applies across the book, and any chapter already produced with the old reading is marked as needing a remix, so a fix made in chapter twelve does not leave chapters one to eleven saying the name the old way. The full workflow, including how to prioritise which words to test, is in our guide to audiobook pronunciation control.
Two honest limits. The first pass is not exhaustive: a rare surname or an unusual spelling can slip past it, so the Add button is still there for the words you know are risky. And the step runs on English-language productions for now; other languages keep the manual list.
Why this matters more for a cast than for a narrator
In a single-narrator recording, a mispronounced name is one voice being consistently wrong, which a listener can at least get used to. In a full-cast production, the same name is spoken by the narrator, by the character who owns it, and by everyone who addresses them. One confirmed reading keeps all of them saying it the same way, and the production keeps its footing. The same list carries the same value across a series, where a name that drifts between books is a continuity failure; our series cast-continuity guide covers that side.
You can hear what a production with a lot of proper nouns sounds like when the names are settled in Frankenstein, where Ingolstadt, Clerval and De Lacey come up often enough that a wrong reading would be hard to miss.
How to try it
Upload a manuscript, open the book's Pronunciation page, and the suggested list is waiting. For a book already in your library that has no pronunciation entries yet, the same page has a Suggest button that runs the pass on demand. Listen to the ones with the highest counts first, use the readings that match your book, and generate.
The narrator was always going to meet your characters' names. Now you get to introduce them.
