How-tos
Add pronunciation aliases — fix how AI voices say names and brands
Pronunciation aliases tell GSpeech how to speak a word or phrase without changing the visible page text. Use them for brand names, people names, acronyms, and technical terms that modern AI voices still misread.
- Spoken fix only — the public HTML and SEO text stay as written.
- Website or widget — set aliases once for a site, or only for one player.
- Language / locale — optional :en or :en-US when the same spelling needs different speech.
- Save applies — no separate regenerate ritual; listen, then hard-reload / cache if you still hear the old form.
Open Cloud Console
Full aliases blog post
All How-tos
Setup
How to add pronunciation aliases
- Open your website in the GSpeech dashboard.
- Open Aliases for the website or for the selected widget.
- Add one alias per line using find:replacement or find:replacement:language.
- Save — aliases apply automatically. No manual regenerate step.
- Listen with the production voice. If you still hear the old pronunciation, hard-reload the page or clear cache.
On WordPress, the cache index updates for you. For custom HTML connection code, bump v_ind (for example ?v_ind=2) so the latest settings load without a hard reload.
Tips
Best practices
- Listen first — modern AI voices already handle most words; alias only real misses.
- One change at a time — easier to compare before/after.
- Keep visible text correct — put speech guidance in aliases, not in the public article.
- Match the active language — the replacement should be readable by the selected voice locale.
- Website vs widget — brand-wide terms at website level; one-off player quirks at widget level.
FAQ
Pronunciation aliases FAQ
Short answers for the Cloud Console workflow.
-
A rule that replaces a written word or phrase with a clearer spoken form before audio is generated. The visible website content stays the same.
-
No. Only the speech pipeline uses the replacement. Readers and search engines still see the original HTML.
-
Yes. Use language-based or locale-specific lines so each active language gets its own spoken form.
-
No separate regenerate step. After Save, the new rules apply. If playback still sounds old, hard-reload or clear cache; on custom embeds, bump v_ind.
-
No. Listen first, then alias only words that genuinely need a correction for your production voice.
Want examples, locale tables, and the full workflow story? Read the deep dive: Language-based pronunciation aliases.
