How to Build a Pronunciation Guide for Character and Place Names

If your book has a character named Ao, and you have already braced yourself for how a voice might read it, you know the problem. Invented names are half the fun of fantasy and science fiction, and they are also the fastest way to hear a narration go sideways. A pronunciation guide fixes that. It is a short, boring document that saves you a lot of re-listening, and it is worth building before you generate a single chapter of audio. Here is how we would put one together.

Why invented names trip up narration

A human narrator asks the author how to say a name, or guesses from context and stays consistent. An AI voice does neither on its own. It reads the letters in front of it and applies the pronunciation rules of the language it was trained on. For ordinary English words that works well. For a name you invented, the voice has nothing to go on except spelling, so "Caedmon" might come out with a hard C, "Xoco" might get an English X, and a place like "Ynys" might be flattened into something you never intended.

The good news is that this is predictable, which means it is fixable. The voice will mispronounce the same name the same way every time, so once you know how a name reads on the page, you can steer it. The lever you control is the text itself. If you tell the voice, through spelling, exactly what sounds you want, it will give them to you consistently across the whole book. That is the whole idea behind a pronunciation guide: decide the right sound once, then make the text carry that decision everywhere the name appears.

Building a pronunciation list before you generate

Start with a simple list, not a system. Open a note or a spreadsheet and write down every proper noun a reader would not already know how to say: character names, place names, invented titles, made-up creatures, spells, house names. Skip the ordinary words. You are only collecting the ones a voice might guess wrong.

Next to each name, write two things: how you actually want it said, in plain description ("KAYD-mon, hard first syllable"), and a respelling that produces that sound (more on that below). Doing this as a batch, before you generate, is much faster than catching mispronunciations one at a time while you listen. It also keeps you consistent. If your protagonist's name shows up four hundred times, you want it decided once, not re-decided every time you hit a chapter. When you build an audiobook the way we describe in our guide to making an audiobook with AI, this list becomes a small prep step that pays for itself immediately.

Spelling names phonetically for the voice

The most reliable way to control an AI voice is to respell the name the way it sounds, using ordinary letters and syllable breaks. You are not changing your manuscript. You are giving the voice a version of the word it can read correctly. A few habits that help:

• Break the name into syllables with hyphens or spaces: "Kae-dmon" or "Kayd mon."
• Use common English spellings for each sound. "Ny-sha" reads more reliably than "Nysha."
• Mark the stressed syllable by capitalizing it: "kayd-MON" versus "KAYD-mon" are two different names.
• Replace letters that carry the wrong rules. A silent or soft consonant is easier to control if you spell around it.
• Keep a single respelling per name and reuse it. Consistency beats cleverness.

Because you are editing text rather than hunting through menus, you can do this in the editor directly. Switch to a clean text view, find the name, and swap in your respelling where you need the voice to follow it. If a name only ever appears in narration and dialogue in the same form, one consistent respelling covers the whole book.

Handling fantasy and non-English names

Two cases need a little extra care. The first is dense invented vocabulary, the kind of world where half the nouns are new. Epic and secondary-world stories live here, and the fix is the same list, just longer. If you are working on something like an epic fantasy audiobook or a science fiction audiobook with invented tech and species names, build the guide as you would a glossary and keep it open while you prep.

The second case is names drawn from real languages other than English: Welsh, Japanese, Nahuatl, Igbo, and so on. Decide early whether you want an authentic pronunciation or an anglicized one, because both are valid choices and readers will notice which you picked. Once you decide, respell toward that target. If your story leans on distinct voices and accents across a cast, our notes on giving each character a distinct voice and accent and on producing a full-cast audiobook pair well with a solid name guide, since voice, accent, and pronunciation together are what make an ensemble sound intentional.

Checking and fixing pronunciations

Do not generate the whole book and hope. Generate a short sample first, a paragraph or two that includes your hardest names, and listen. This is the single most useful habit in the whole process, and it costs almost nothing because you are only voicing a few lines. Any generation, including re-generation, counts against your monthly word allowance, so finalizing the tricky names on a sample keeps a full-book pass from being an expensive guess. On the free tier you get a small monthly allowance with no card, which is plenty for testing names before you commit.

When a name still comes out wrong, adjust the respelling and generate that short sample again. Two or three passes usually settle even a stubborn name. Once the sample sounds right, apply the same respelling everywhere the name appears and generate the full chapters. When you are done, you download a finished MP3. We produce the file; you take it and publish it wherever you already publish, whether that is a retail store, your own site, or a supporter feed. The pronunciation guide travels with you into every future book in the series, so the second audiobook starts from a list that already works.