Pronunciation settings define how text-to-speech should say specific words or phrases. They are useful for character names, invented terms, uncommon spellings, and words with different spoken forms.
Use a simple written replacement when that is enough. Use provider-specific pronunciation values only when the target voice requires them and the audio sample confirms the result.
New rules match a Word or phrase literally. Whole words is on and Match case is off by default. The Regular expression checkbox is beside the word/pattern label. Say it as and Phonetic symbols appear alongside each other. Existing regex rules remain regex rules. Advanced users only. Regular expressions are complex: a broad or incorrect pattern can change unintended words throughout the prepared audio. Leave Regular expression unchecked for ordinary names and phrases.
Pronunciation patterns use Python regex, not the JavaScript replacement syntax used by manuscript Find. \b marks a word boundary, \s matches whitespace, \. matches a literal period, and ( ... ) captures matching text (omit the example spaces). For example, pattern \bDr\.(\s) with Say it as Doctor\1 turns Dr. Smith into Doctor Smith; \1 reuses the captured space. Pattern \bcolou?r\b matches color or colour because ? makes the preceding character optional. Matching is case-sensitive unless the pattern uses a flag such as (?i). Avoid broad patterns such as .*. Test matching and nonmatching sentences with Test matches before saving changes, then listen to a short preview. These rules affect narration output, not manuscript spelling.
Choose Edit / test and a preview voice. Enter a test sentence, then Test matches to see the result without a provider charge. Listen prepares that sentence using all effective draft rules. Replays reuse the completed audio; changing the rules, sentence, or voice requires a new test. Audio previews use the normal funding, quota, and job workflow.
Phonetic symbols take precedence for compatible voices: xAI supports IPA, and Kokoro supports its phonetic alphabet. Text-only adapters use Say it as instead. If no text fallback is set, the original spelling is retained and a warning appears. Effective method identifies which value the selected voice uses.
Drag pronunciation rows or use their up/down buttons. The first row runs first, followed by each row beneath it. Put more specific or protective patterns above broader patterns when rules can overlap.
Open Project library > Pronunciations, directly below Reference and Guidance and before Images. Rules apply to the entire project, with no book or manuscript overrides. Save changes persists immediately. Drag rows or use the up/down buttons to save their order.
Select manuscript text and choose Add pronunciation to open a literal project-wide rule over the manuscript. Save changes saves directly without leaving the editor; closing restores its focus and position. The current book supplies preview voice context only. Viewing is available without project editing permission; saving requires it.
Choose Listen for snippets, selected blocks, voices, pronunciations, and effects. In Listen to selection: Audio starts automatically when it is ready. Use Stop to end playback. If your browser blocks automatic playback, choose Play to start it manually. You can also download the prepared audio. Check a short selection before publishing the entire manuscript.
Publish manuscript prepares files for the entire manuscript, even when you select only audio. Choose the output formats, then Publish. Publication ready means the files are ready to download, not that your work has been distributed publicly.