VOICE AI PRONUNCIATION WORKSHEET
Free resource. Speech synthesis, platform usage and carrier costs are separate.
Prepared October 6, 2026.

A. Identify the failing word
Exact approved spelling:
Desired spoken form, approved by the business owner or a fluent speaker:
Target language / locale:
Voice provider / exact model / voice ID:
Sentence where the problem occurs:
Does the response text already contain the correct word? yes/no:
If no, fix source facts or response generation before changing TTS.

B. Fictional worked example
Business: Wexley Studio (fictional).
Canonical name: Wexley Studio.
Business-approved spoken name: WEKS lee Studio (illustrative).
Response text: 'Thanks for calling Wexley Studio. How can I help?'
Repeated phrase: 'Your enquiry for Wexley Studio needs a staff callback.'
Create a provider pronunciation rule for Wexley using its documented format.
Keep Wexley Studio as the canonical name in business facts and saved records.
A sounds-like spelling is guidance, not an automatically verified pronunciation.

C. Cartesia connection checklist
Create the dictionary in your authorized Cartesia account using its current documentation.
Use the same account credentials for synthesis, or resolve dictionary access with the owner.
Keep private business/customer vocabulary private; do not share a dictionary just to bypass access.
Record dictionary ID and actual revision/export:
Burki Voice: choose Standard pipeline under Conversation engine.
If the engine control is collapsed, expand Model and provider choices.
Then Speech settings > Advanced voice stack > Cartesia.
Choose a supported Sonic 3 model and voice.
Open Voice tuning > Pronunciation Dictionary ID and use the actual dictionary ID.
Review applicable provider credentials, billing selection and funding before a test.
Do not paste the dictionary ID or IPA into Conversation as though it were a provider setting.
For GPT Live, the configured TTS provider is the announcement voice, not its conversation voice.

D. Acceptance sheet
Phrase / language / model / voice / dictionary ID / observed audio / result / correction:
Greeting with the name:
Name in a longer sentence:
Name next to a number or address:
Capitalization variants actually used by the assistant:
An unrelated word that must remain unchanged:
Correction or interruption that triggers the phrase again:

Check exact model support for case handling, phonemes and markup.
Repeatedly listen to changed terms before relying on them.
Save the actual observed pronunciation or leave it untested.
There is no synthetic recording or customer success claim in this resource.