Vois
Back to BlogTips & Tricks

The Pronunciation Dictionary That Makes AI Voices Sound Human

Vois TeamVois Team
February 27, 2026
6 min read

TLDR:Vois includes a pronunciation dictionary that lets you define how specific words are spoken. Add a rule once ('Vois' becomes 'Voyce') and it applies everywhere. Supports global and project-specific scopes, CSV import/export for bulk management, and integrates into the text normalization pipeline automatically.

A nearly perfect voiceover can lose credibility when it says a product, name, or acronym wrong. Vois's pronunciation dictionary gives you a repeatable way to catch that error before it reaches an audience.

Or it pronounces your company name like it's never heard of it. Or it reads "LUFS" as a word instead of spelling out the letters. Or it turns "Dr. Mehta" into something unrecognizable.

Every voice system has blind spots. Names, brands, acronyms, technical terminology, and foreign words in English context can break an otherwise natural delivery.

The Vois pronunciation dictionary lets you make an approved spoken form part of the workflow, rather than correcting the same word in every script.

Person writing and editing

How It Works

The concept is simple: you tell Vois how a word should be pronounced, and it remembers.

Add a rule: "Vois" should be spoken as "Voyce." From that point forward, every time your text contains "Vois," the engine sees "Voyce" instead. The substitution happens before audio generation, automatically every time.

Rules are stored in a local database. They persist across sessions, across projects, across restarts. Add a rule once and forget about it.

Words That Need Rules

Certain categories of words reliably cause problems:

Brand Names

Most brand names are invented words with non-obvious pronunciation. Voice engines have to guess, and the guess can be wrong.

Brand TTS Might Say Rule
Vois "Voh-iss" "Voyce"
Canva "Can-vah" "Can-vuh"
Asana "Ah-sah-nah" "Ah-sahn-uh"
Figma "Fig-mah" "Fig-muh"

If you're producing content for a specific brand, add their name on day one. It'll save you from catching the mispronunciation after you've already generated 20 minutes of audio.

Character Names in Fiction

Audiobook authors deal with this constantly. Fantasy and sci-fi names are particularly tricky:

Name Problem Rule
Aethon "Ay-thon" when you want "Ee-thon" "Ee-thon"
Cthulhu Anybody's guess "Kuh-too-loo"
Hermione Often mangled "Her-my-oh-nee"

For audiobooks with large casts, you might have dozens of character names that need rules. This is where CSV import becomes essential (more on that below).

Acronyms and Initialisms

TTS engines have to decide: is this an acronym (read as a word) or an initialism (spell out the letters)?

Term Common Error Rule
LUFS Read as "luffs" "L U F S" (spell it out)
ASMR Read as "azmer" "A S M R"
SQL Read as "sequel" or "S Q L" Your preference
API Usually correct Verify first
GDPR Sometimes "guh-dpr" "G D P R"

The space-separated letters trick works well: writing "L U F S" in the replacement forces the engine to spell it out letter by letter.

Medical and Legal Terminology

Specialized fields have vocabulary that general TTS models rarely encounter in training data:

Term Common Error
Dysphagia Stress on wrong syllable
Voir dire Anglicized incorrectly
Amicus curiae Butchered Latin
Ophthalmology Dropped syllable

If you're producing medical or legal content regularly, build a domain-specific dictionary. It's an investment that pays off across every future project.

Tools and configuration

Global vs Project-Specific Rules

Rules come in two scopes:

Global rules apply across every project. Use these for permanent vocabulary, such as your brand name, common acronyms, and frequently used technical terms. These are your "set and forget" rules.

Project-specific rules apply only within a single project. Use these for context-dependent pronunciation, such as character names in a specific audiobook, product names in a specific training series, or jargon relevant to one client.

Most people start with global rules for their most common terms, then add project-specific rules as needed. The two layers do not conflict. Project rules take priority over global rules when both exist for the same word.

The CSV Workflow

Adding rules one at a time works for a handful of terms. If you are building a comprehensive dictionary, such as an audiobook with 40 character names or a medical course with 100 technical terms, use bulk import.

Export your current dictionary as CSV, add new entries in a spreadsheet, and import the updated file. The CSV format is two columns: original word and replacement.

original,replacement
Vois,Voyce
LUFS,L U F S
ASMR,A S M R
Mehta,May-tah

This also makes dictionaries portable. Export from one Vois installation, share the CSV with a colleague, import on their machine. Consistent pronunciation across your team.

Word Matching Details

The dictionary uses whole-word matching. A rule for "Dr" replaces "Dr" but not "Dr" inside "Dragon" or "Dry." This prevents the kind of cascading substitution errors that make dictionary systems unreliable.

The matching algorithm processes longest matches first. If you have rules for both "New York" and "York," the two-word phrase gets priority. This lets you create compound rules without worrying about partial matches breaking them.

Achievement and success

Building Your Dictionary Over Time

The best pronunciation dictionaries aren't built in a day. They grow organically:

  1. Start with known problems. Your brand name, your host's name, your most common acronyms. Add these before your first generation.

  2. Catch issues during review. When you listen to generated audio and hear a mispronunciation, add a rule immediately. Do not postpone it until the term is easy to forget.

  3. Import domain vocabularies. If you're entering a new content domain (medical training, legal content, a new game universe), build a starter dictionary from your terminology list before generating anything.

  4. Export and backup. Periodically export your dictionary as CSV. It represents accumulated knowledge about how your content should sound. Treat it like any other production asset.

The Difference It Makes

A pronunciation dictionary is a quiet production feature, but it makes the difference between audio that is almost right, with jarring mispronunciations breaking the flow, and audio that sounds consistently professional.

The first time your AI voice nails a difficult name that it used to butcher, you'll understand why this feature exists.

Set the rule once, review the rendered line, and carry the approved pronunciation into the next project. Explore the Vois pronunciation dictionary, then get started when you are ready to keep your vocabulary consistent.

The Vois Team

Frequently Asked Questions

How does the pronunciation dictionary work?

You add rules mapping written words to phonetic alternatives. When Vois processes text, it substitutes the written form with your pronunciation before generating audio. Rules apply automatically to all future generations.

Can I import pronunciation rules in bulk?

Yes. Vois supports CSV import and export for pronunciation rules. Create a spreadsheet with two columns (original and replacement), export as CSV, and import the entire vocabulary at once.

Does the pronunciation dictionary work across all projects?

Yes. Rules can be set as global (applying across all projects) or project-specific. Global rules are the most common choice for brand names, acronyms, and frequently used terms.

What kinds of words need pronunciation rules?

Common candidates include brand names, character names in fiction, technical acronyms, medical or legal terminology, foreign words used in English context, and any word the TTS engine consistently mispronounces.

PronunciationTipsAi VoicesProductionTutorials
Share:
Vois Team

Written by

Vois Team

Product Team

The team behind Vois, building the future of AI voice production.