A pronunciation guide tells a narrator what to say when a script contains unfamiliar names, places, initials, or technical terms. AI can find likely trouble spots and organize a first pass. Keep each pronunciation traceable to a person, a source, or an approved creative decision rather than asking the model to guess.
Find the terms that could slow down the read
Give each scene or paragraph a stable label, such as S03 or P12. Ask the model to list potentially tricky terms exactly as written, with their first location and a reason for flagging them. Look for names with several plausible readings, stylized product names, initials, and invented words whose stress matters. Share only material your approved tool is allowed to handle; a short excerpt is often enough.
Give each cue a source and a status
Use a person’s own spoken introduction for their name when available, a recording from the organization for its name, or a reputable dictionary with audio for an ordinary term. For a fictional name, ask the author or production lead. Record the relevant timestamp or written approval.
- Confirmed: You checked a reliable reference that applies to this exact use.
- Needs check: The reference is missing, ambiguous, or disagrees with another source.
- Production choice: An authorized person chose a reading for the character or performance; record who and which version.
Write cues the narrator can scan
Keep the script spelling untouched. Add the cue in a separate guide or margin. Capitalize the stressed syllable if you use plain-language cues. For initials, say whether to read each letter or pronounce a single word. Use an audio link for a sound that written cues don’t convey clearly.
A fictional example
Scene S03 opens: “The Avenor Signal reached Vey on the RST relay.” The terms below are made up for this example.
- Avenor | S03 | AY-vuh-nor; first-syllable stress | Producer approval, version 2 | Production choice.
- RST | S03 | Say R, S, T separately | Producer’s read-through | Confirmed for this script.
- Vey | S03, S09 | No cue yet | Ask the author for a short recording | Needs check.
Leave the unresolved line for a pickup or have the production lead document a temporary choice. Don’t silently replace a missing answer with the model’s best guess.
Reuse this prompt
Build a pronunciation guide from my labeled script. List difficult names, places, initials, and terms exactly as written.
Return: Term | First scene | Read-aloud cue | Evidence | Status | Next step.
Use my supplied references only. Do not invent a cue, source, timestamp, or approval.
Use Confirmed, Needs check, or Production choice. Leave the cue blank when evidence is missing and flag disagreements instead of choosing between them.
Script: [labeled excerpt]
Reference notes: [checked audio timestamps, dictionary entries, or author decisions]
Check each term in the actual sentence
Record a short read with the intended narrator or speech voice. Compare flagged terms against the checked references at the natural pace of the sentence. Listen for the stressed syllable, separated initials, and consistent readings later in the script. For synthetic speech, use the actual voice and settings; the same spelling cue may sound different in another voice.
Save the guide against the current script version. Before recording, scan newly added lines and resolve any outstanding entries or explicitly mark them for pickup.

