Build the pronunciation list in four steps
- 1
Collect the hard words
Skim the manuscript for character names, places, invented terms, foreign phrases and technical words. Searching for capitalized words that are not at the start of a sentence finds most of them.
- 2
Decide how each one sounds
Say each word aloud and settle on one version. For real names and places, check how locals or the person themselves say it rather than guessing.
- 3
Write a respelling
Spell the sound with ordinary English syllables, and mark the stressed syllable in your list. Keep the original spelling next to it.
- 4
Test it in a voice
Paste one sentence containing the word into the AI audiobook generator and listen in the narrator’s voice. Adjust the respelling until it lands, then move to the next word.
What entries in a pronunciation guide look like
A guide is a plain table. Capital letters mark the stressed syllable for whoever reads the list.
| Word | Respelling | Notes |
|---|---|---|
| Siobhan | shi-VAWN | Irish given name |
| Aoife | EE-fa | Irish given name |
| Worcester | WUSS-ter | English city; never “war-chester” |
| Hermione | her-MY-oh-nee | Four syllables |
| Kaeldrith | KAYL-drith | Invented city; stress the first syllable |
| Lyrae | LEER-ay | Invented name; two syllables, not “liar” |
| Qiao | chee-OW | Mandarin surname; the Q sounds like ch |
Fantasy, sci-fi and cultivation novels need the longest lists. See fantasy audiobook narration and xianxia audiobook for genre casting tips.
How to write a phonetic respelling a voice will follow
Use everyday spellings for sounds. “Ay” as in day, “ee” as in see, “igh” as in high, “oo” as in moon, “ow” as in cow, “zh” as in measure. A voice trained on English text reads these predictably. Phonetic alphabet symbols such as ə or ʃ, by contrast, are usually skipped or read oddly when they appear in plain text.
Keep capitals in the list, not in the text. Capitals are the clearest way to show stress to a person, but a voice may read a capitalized syllable as letters, so “shi-VAWN” can come out spelled. In the text you generate, write the respelling in lowercase, such as “shivawn” or “shi-vawn”, and steer stress by how you split the syllables.
Split where the word should break. If a voice puts the stress in the wrong place, move the break or double a letter. An invented name listed as “ka-LEEN-a” might go into the text as “kaleena”, then “ka-leena”, then “kah-leena” until one sounds right. Two or three tries is normal.
Change the copy, not the book. Respellings go into the file you narrate, not the ebook you sell. Keep the narration copy separate, and use find and replace so every occurrence gets the same respelling.
Words spelled one way and said two ways
Heteronyms trip up every voice now and then because the right reading depends on meaning. Context usually solves it; when it does not, these fixes do.
| Word | Readings | Fix in the narration copy |
|---|---|---|
| read | reed (now), red (past) | Write “red” where the past tense is misread |
| lead | leed (guide), led (metal) | Write “led” for the metal if needed |
| tear | tair (rip), teer (cry) | Rephrase, or respell as “tair” or “teer” |
| wind | wind (air), wynd (turn) | Respell the verb as “wynd” |
| live | liv (verb), lyve (adjective) | Respell the adjective as “lyve” |
| bass | bayss (music), bass (fish) | Respell the music sense as “base” |
| minute | MIN-it (time), my-NOOT (tiny) | Write “my-noot” for the adjective |
Fixing one misread name without redoing the chapter
Generate chapter one first and listen with the text in front of you, noting every name that sounds wrong. Because each line is its own clip, you edit the text of that one line, regenerate it and the rest of the chapter stays exactly as it was.
Then carry the fix forward. Apply the same respelling with find and replace in your narration copy before you generate the remaining chapters, so the name is right everywhere on the first pass instead of being fixed line by line later.
AI voice pronunciation varies from voice to voice, so test names in more than one. The narrator, the main characters and the different voice models (Standard, Lite, Premium and Expressive) can each read the same respelling slightly differently. A name that is spoken mostly in dialogue should be tested in the voices of the characters who say it most. Browse the cast options in AI voices for audiobooks.
Foreign phrases, accents and real places
An English voice reading a French phrase will tend to anglicize it. For a single word that is often acceptable, and listeners expect it. For a phrase that matters, respell it as an English speaker would hear it, or rephrase so the meaning comes through in English. Where a character speaks whole lines in another language, the voice library covers 13 languages, so try a voice native to that language on one line and keep it if it sounds right.
Real places carry local rules that dictionaries miss: Houston Street in New York is HOW-stun, Greenwich is GREN-itch, and a town in your own region may have its own quirk. Check audio sources such as Forvo, YouGlish or local news clips. For living people, the person’s own pronunciation always wins.
Accents themselves belong to casting, not spelling. Writing dialect phonetically into dialogue usually makes a voice sound strained. Choose a voice with the accent you want instead; how to do character voices covers when an accent helps and when it distracts.
Keeping audiobook pronunciation consistent across a series
- Keep one master list per series, stored with your character notes, and add to it with every book.
- Reuse the same cast, so the narrator and recurring characters sound the same in book two as in book one.
- Reuse the same respellings, even if you would choose a different one today. Listeners notice a name that changes between books.
- Note where each word first appears, by chapter, so it is easy to check when a new book refers back to it.
- Prepare the list before the manuscript, using the same steps as the audiobook script format guide.
If a human narrator reads the book instead
The same list works as a pronunciation guide for narrators. Add the chapter of first appearance and a short voice memo of you saying each word, and send it before recording starts. Narrators often charge for pickups caused by names that changed after recording, so settle every name first.
