Voice AI Pronunciation Guide
This guide focuses on optimizing the way Voice AI pronounces brand names, contact names, and industry-specific jargon using the standard text and preview box. Since the agent processes text naturally, the best way to correct mispronunciations is through Phonetic Respelling. This involves changing the spelling of a word in your script to a version that “sounds like” the intended result.1. The Phonetic Respelling Strategy
When the Voice AI mispronounces a word, replace it in your script with a phonetic version. Use hyphens to separate syllables and capitalization to indicate which syllable the agent should emphasize.2. Best Practices for Clear Speech
To get the most natural sound out of the preview box, follow these formatting rules:- Break it down: Use hyphens for multi-syllable names (e.g.,
Me-ganvsMay-gan). - Use Emphasis: Capitalize the “stressed” part of the word. A voice agent will naturally lift the pitch and volume of capitalized letters.
- Example:
pro-DUCE(the verb) vs.PRO-duce(the fruit).
- Example:
- Check for Homonyms: Some words are spelled the same but sound different. If the agent says “I read (red) the book” when you want “I will read (reed) the book,” change the spelling to
reed.
3. Dealing with Acronyms and Numbers
Voice AI sometimes struggles to decide whether to say a word or spell it out.- To force spelling it out: Use periods or spaces between letters.
- Example:
T.H.R.E.A.D.orT H R E A D.
- Example:
- To force a word: Spell it as it sounds.
- Example: For the “SQL” database, use
Se-quelinstead ofS.Q.L.
- Example: For the “SQL” database, use
- Dates and Years: If “2024” sounds robotic, try typing
twenty twenty-four.
4. The “Trial and Error” Workflow
Because you have access to the Preview Box, use this 3-step loop to perfect your brand voice:1
Paste & Listen
Paste your actual script into the box and generate the audio.
2
Identify "Glitch" Words
Note any words where the agent sounds robotic, hesitant, or incorrect.
3
Swap & Test
Replace only those specific words with phonetic versions (e.g., changing
Getthread to Get-thred) and preview again until it sounds human.