Elevenlabs Pronunciation Dictionary: Complete 2026 Guide

If you have ever generated speech with ElevenLabs and heard it mangle a brand name, acronym, or technical term, the ElevenLabs pronunciation dictionary is the fix. Instead of rewording your text to trick the model into saying things correctly, you define explicit pronunciation rules that apply every time a word appears - across every voice and every generation.

This guide walks through the complete setup process for pronunciation dictionaries, covering both rule types, real-world examples for common problem words, and a testing workflow so you can validate corrections before they hit production audio. The entire configuration takes about 20 minutes, and dictionaries are available on all ElevenLabs plans including the free tier.

Prerequisites

Before creating your first pronunciation dictionary, make sure you have the following ready.

An active ElevenLabs account. The ElevenLabs pronunciation dictionary is available on every plan, including free - though you can confirm tier coverage on the ElevenLabs pricing page. If you have not signed up yet, create an account at ElevenLabs and familiarize yourself with the basic text-to-speech workflow first. Our Getting Started with ElevenLabs guide covers account creation and your first generation, and the official ElevenLabs documentation has reference material for every API endpoint.

A list of problem words. Before diving into the dictionary editor, spend five minutes writing down every word that ElevenLabs mispronounces in your content. Common categories include brand names (Figma, Canva, Vercel), technical acronyms (YAML, SQL, OAuth), product names (iPhone, ChatGPT), and foreign-origin words used in English text. Having this list ready makes the setup process much faster.

Basic familiarity with the ElevenLabs interface. You should know how to generate speech from text using the Speech Synthesis page. If you have used ElevenLabs Studio for longer projects, that works too - dictionaries apply in both contexts.

What Are the ElevenLabs Pronunciation Dictionary Rule Types?

The ElevenLabs pronunciation dictionary supports two distinct rule types, each designed for different correction scenarios. Understanding when to use each one is the key to getting reliable results - and most production setups end up using both inside the same ElevenLabs pronunciation dictionary.

Phoneme Rules (English Only)

Phoneme rules let you supply ElevenLabs phonemes through the International Phonetic Alphabet (IPA) wrapped in PLS XML format to specify exactly how a word should sound. Because ElevenLabs IPA notation captures the precise sequence of sounds, you leave no room for interpretation. This is the most accurate correction method available.

When to use phoneme rules:

• Words where the model consistently gets the stress or vowel sounds wrong
• Technical terms with non-standard pronunciation
• Names that follow unusual phonetic patterns
• Any word where you need exact, repeatable pronunciation

Limitation: Phoneme rules only work for English-language content. If you are generating speech in other languages, you need alias rules instead.

Alias Rules (All Languages)

Alias rules work by replacing one word with another before the model processes the text. Instead of defining phonetic sounds, you provide a substitute word or phrase that the model already pronounces correctly. The replacement happens silently - the original word appears in your text, but the model reads the alias.

When to use alias rules:

• Multilingual content where phoneme rules are not supported
• Words that sound like common English words (just spelled differently)
• Acronyms you want expanded into full words
• Quick fixes where you do not need phonetic precision

Advantage: Alias rules work across all 32 supported languages, so they are the go-to option for multilingual projects and not just an ElevenLabs pronunciation dictionary English-only fix.

How Do You Create Your First Pronunciation Dictionary?

Navigate to the ElevenLabs dashboard and open the Pronunciation Dictionaries section from the left sidebar. Click Create to start a new dictionary.

Step 1: Name your dictionary. Choose a descriptive name that reflects its purpose. If you are creating dictionaries for different projects or clients, include that context in the name. For example, “SaaS Product Names” or “Medical Terminology - Patient Portal.” You can create multiple dictionaries and apply different ones to different voices or projects.

Step 2: Add your first rule. Click Add Rule and you will see two options: Phoneme and Alias. Start with whichever type matches your first problem word. The following sections cover both in detail.

Step 3: Test the rule. After adding a rule, use the built-in preview to generate speech with a sentence containing the target word. Compare the output against a generation without the dictionary applied. If the pronunciation is correct, move on to the next word. If not, adjust the rule.

Step 4: Apply the dictionary to a voice. Once your dictionary contains all the rules you need, assign it to one or more voices. Navigate to the voice settings and select your dictionary from the pronunciation options. Every generation using that voice will now respect your pronunciation rules automatically.

Writing Phoneme Rules with PLS XML

Phoneme rules follow the Pronunciation Lexicon Specification (PLS) format, an XML standard that maps written words to their phonetic representation. Here is the basic structure.

<?xml version="1.0" encoding="UTF-8"?>
<lexicon version="1.0"
  xmlns="http://www.w3.org/2005/01/pronunciation-lexicon"
  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
  xsi:schemaLocation="http://www.w3.org/2005/01/pronunciation-lexicon
    http://www.w3.org/TR/2007/CR-pronunciation-lexicon-20071212/pls.xsd"
  alphabet="ipa" xml:lang="en-US">

  <lexeme>
    <grapheme>Figma</grapheme>
    <phoneme>fIgm@</phoneme>
  </lexeme>

</lexicon>

The critical elements are:

• <grapheme> - The word as it appears in your text (case-sensitive matching)
• <phoneme> - The IPA transcription defining how the word should sound
• alphabet="ipa" - Tells the system you are using IPA notation
• xml:lang="en-US" - Specifies the language context (English only for phoneme rules)

IPA Phoneme Reference for Common Sounds

Getting IPA notation right is the hardest part of phoneme rules. Here is a practical reference chart for the sounds you will use most often when correcting brand names and technical terms.

Real-World Brand Name Corrections

Here are PLS XML examples for brand names that ElevenLabs commonly mispronounces. You can copy these directly into your dictionary.

<?xml version="1.0" encoding="UTF-8"?>
<lexicon version="1.0"
  xmlns="http://www.w3.org/2005/01/pronunciation-lexicon"
  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
  xsi:schemaLocation="http://www.w3.org/2005/01/pronunciation-lexicon
    http://www.w3.org/TR/2007/CR-pronunciation-lexicon-20071212/pls.xsd"
  alphabet="ipa" xml:lang="en-US">

  <lexeme>
    <grapheme>Figma</grapheme>
    <phoneme>fIgm@</phoneme>
  </lexeme>

  <lexeme>
    <grapheme>Canva</grapheme>
    <phoneme>kaenv@</phoneme>
  </lexeme>

  <lexeme>
    <grapheme>Vercel</grapheme>
    <phoneme>v3:rsel</phoneme>
  </lexeme>

  <lexeme>
    <grapheme>Astro</grapheme>
    <phoneme>aestr@U</phoneme>
  </lexeme>

  <lexeme>
    <grapheme>Supabase</grapheme>
    <phoneme>su:p@beIs</phoneme>
  </lexeme>

  <lexeme>
    <grapheme>Tailwind</grapheme>
    <phoneme>teIlwInd</phoneme>
  </lexeme>

</lexicon>

Tip: When you are unsure of the correct IPA for a word, search for the word on Wiktionary, which provides IPA transcriptions for most English words. For brand names, check the company’s own pronunciation guides or promotional videos to confirm the intended pronunciation before writing your rule. The International Phonetic Association publishes a full IPA chart if you want a complete reference.

Writing Alias Rules

Alias rules are simpler to write because they do not require IPA knowledge. You define a source word and a replacement word that the model already knows how to pronounce.

Basic Alias Structure

In the dictionary editor, select Alias as the rule type and fill in two fields:

• Original word: The word as it appears in your text
• Alias: The replacement word or phrase the model should read instead

Here are common alias corrections:

When Aliases Fall Short

Alias rules have a limitation - the replacement word must already be pronounced correctly by the model. If both the original and the alias are uncommon words, the model might mispronounce the alias too. In those cases, you need phoneme rules for English content or need to find a more common-sounding alias.

For example, aliasing “Netlify” to “net-lih-fy” works because each syllable is a recognizable English sound. But aliasing a foreign name to another foreign name might not improve anything.

Multilingual Pronunciation

For content in languages other than English, alias rules are your only option - but they work well across all 32 languages ElevenLabs supports. The approach changes slightly depending on the language context.

Same-Language Aliases

When generating speech in French and need to correct a word’s pronunciation within French, provide a French alias. The model processes the alias in the same language context, so the replacement needs to make phonetic sense in that language.

Cross-Language Brand Names

The most common multilingual use case is brand names appearing in non-English content. A German-language podcast mentioning “Stripe” might hear the model pronounce it with German phonetics. Create an alias that guides the model toward the English pronunciation while keeping it natural within the target language’s sound system.

Dictionary Per Language Strategy

If you produce content in multiple languages, create separate dictionaries for each language rather than mixing rules in one dictionary. Name them clearly - “Brand Names - English,” “Brand Names - German,” and so on. This keeps rules organized and avoids conflicts where a correction that works in one language causes problems in another.

Apply the appropriate dictionary when switching between voices or language contexts in your projects. In ElevenLabs Studio, you can assign different dictionaries to different speakers within the same project.

Testing and Validation Workflow

A pronunciation dictionary is only useful if every rule actually works. Rushing through testing leads to embarrassing audio where “fixed” words sound worse than the original mispronunciation. Follow this structured testing process.

Step 1: Create a Test Script

Write a short paragraph that uses every word in your dictionary at least once, ideally in natural sentence contexts rather than a bare word list. The model’s pronunciation can shift depending on surrounding words, so testing in isolation is not enough.

Example test script for the SaaS brand dictionary:

“We migrated our Figma designs to a Vercel-hosted Astro site with a Supabase backend. The Tailwind CSS framework and Canva graphics made the redesign straightforward. Stripe handles our payments.”

Step 2: Generate Without the Dictionary

Run the test script through speech synthesis without the pronunciation dictionary applied. Save or note the output so you have a baseline for comparison.

Step 3: Generate With the Dictionary

Apply your dictionary and run the same script again. Listen carefully to each target word. Check for:

• Correct pronunciation - Does the word sound right?
• Natural flow - Does the corrected word fit naturally in the sentence, or does it sound jarring?
• Stress patterns - Is the emphasis on the right syllable?
• Speed consistency - Does the model slow down or speed up unnaturally around the corrected word?

Step 4: Test Across Multiple Voices

A rule that works with one voice might behave differently with another. Test your dictionary against at least two or three voices you plan to use. The Eleven Multilingual v2 and Eleven Turbo v2.5 models can handle pronunciation rules differently, so check across models as well.

Step 5: Iterate

If a rule sounds wrong, adjust it and test again. For phoneme rules, small changes to a single IPA symbol can make a big difference. For alias rules, try different phonetic spellings until you find one that sounds natural.

Applying Dictionaries to Voices and Projects

Once your dictionary passes testing, you need to connect it to the voices and projects that should use it.

Speech Synthesis Page

On the Speech Synthesis page, look for the pronunciation dictionary option in the settings panel. Select your dictionary from the dropdown. Every generation from this page will use those rules until you remove or change the dictionary.

Studio Projects

In ElevenLabs Studio, pronunciation dictionaries can be applied at the project level. Open your project settings and assign a dictionary. This is particularly useful for long-form content like audiobooks, course narration, or podcast series where consistency across multiple chapters or episodes matters.

API Integration

If you use the ElevenLabs API for programmatic speech generation, include the pronunciation dictionary ID in your API calls. The dictionary rules apply server-side, so you do not need to preprocess your text.

<pronunciation_dictionary_locators>
  <locator>
    <pronunciation_dictionary_id>your-dictionary-id</pronunciation_dictionary_id>
    <version_id>your-version-id</version_id>
  </locator>
</pronunciation_dictionary_locators>

Retrieve your dictionary ID and version ID from the dictionary settings page or via the API’s list dictionaries endpoint.

Common Pronunciation Problems and Fixes

Here is a quick-reference table for the types of pronunciation issues you will encounter most frequently, along with the recommended fix approach.

Handling Edge Cases

Words with multiple valid pronunciations. Some words are legitimately pronounced differently in different contexts. “Read” (present) versus “read” (past) is a classic example. Pronunciation dictionaries apply the same rule regardless of context, so you cannot handle these cases with a dictionary alone. Instead, use alias rules with context-specific workarounds - write “reed” in your source text when you mean present tense, and create an alias from “reed” back to the correct output.

Hyphenated compounds. Words like “real-time” and “e-commerce” can trip up the model at the hyphen boundary. If the model pauses awkwardly or shifts stress across the hyphen, create an alias that removes the hyphen or replaces it with a space.

Numbers and symbols. The model generally handles numbers well, but edge cases like “802.11ac” or “3GPP” need aliases. Map them to their spoken equivalents: “eight-oh-two-dot-eleven-ay-see” and “three-G-P-P.”

Best Practices for ElevenLabs Pronunciation Dictionary Management

Start small and expand gradually. Do not try to fix every possible mispronunciation in your ElevenLabs pronunciation dictionary in one session. Begin with the words that appear most frequently in your content and add new rules as you encounter problems. The ElevenLabs developer documentation covers programmatic management for teams that prefer config-as-code.

Document your rules. Keep a separate spreadsheet or document that lists every rule, why you added it, and what it fixes. When you come back to update the dictionary months later, this documentation saves time. Storing this alongside your project files in a tool like GitHub keeps everything version-controlled together.

Version your dictionaries. ElevenLabs supports dictionary versioning. When you make significant changes, create a new version rather than editing rules in place. This lets you roll back if a batch of changes introduces problems.

Review periodically. Models improve over time. A word that needed a pronunciation fix six months ago might be handled correctly by a newer model version. Test your rules against current model behavior quarterly and remove any that are no longer necessary. The ElevenLabs changelog is a useful reference for spotting model upgrades.

Share across teams. If multiple people in your organization use ElevenLabs, share pronunciation dictionaries so everyone benefits from the same corrections. This is especially important for brand name consistency in corporate content - and pairs well with shared workspace setups described in the ElevenLabs Team Workspace Guide.

Frequently Asked Questions

How many rules can a single ElevenLabs pronunciation dictionary contain?

ElevenLabs does not publish a hard limit on rules per dictionary, and in practice, dictionaries with hundreds of rules work without performance issues. The rules are processed at generation time as a lookup table, so adding more entries does not noticeably increase generation latency. That said, organizing rules into multiple focused dictionaries (one for brand names, one for medical terms, one for acronyms) is easier to maintain than a single massive dictionary - especially when several teammates are editing the same ElevenLabs pronunciation dictionary.

Do pronunciation dictionaries work with voice cloning?

Yes. Pronunciation dictionaries work with all voice types - pre-made voices, cloned voices (both instant and professional), and voices created through the Voice Design feature. The dictionary rules operate at the text processing layer, before the voice model generates audio, so they are voice-agnostic. Apply the same dictionary to any voice and the pronunciation corrections will take effect. For details on setting up cloned voices, see our ElevenLabs voice cloning tutorial.

Can I import and export pronunciation dictionaries?

You can add rules via PLS XML files, which makes it straightforward to prepare rules in a text editor and import them in bulk. For teams that manage pronunciation across multiple tools, this XML-based approach means you can version-control your dictionary files in Git alongside your content. Export functionality lets you back up your rules or transfer them between accounts.

What happens when a word matches both a phoneme rule and an alias rule?

If the same word has both rule types defined in a dictionary, the phoneme rule takes priority for English content. In practice, you should avoid duplicate entries for the same word. Use phoneme rules when you need precise control over sounds and alias rules when a simple word substitution is sufficient. If you are unsure which to use, start with an alias rule - it is faster to set up - and switch to a phoneme rule only if the alias does not produce accurate enough results.

Do dictionaries affect generation speed or quota usage?

No. Applying a pronunciation dictionary does not consume additional character credits and does not add measurable latency to generation. The lookup happens before the model processes your text, and the cost is based on the final text length (with alias replacements applied). A 1,000-character input with a dictionary applied costs the same as 1,000 characters without one.

Related Guides

• Getting Started with ElevenLabs - Account creation and first generation
• ElevenLabs Studio First Project - Long-form audio projects with multiple speakers
• ElevenLabs Voice Library Guide - Pick the right voice before applying your dictionary
• ElevenLabs Voice Cloning Tutorial - Create custom voices that respect your pronunciation rules
• ElevenLabs API Developer Setup - Pass dictionary IDs through programmatic calls

Related Reading

• Getting Started with ElevenLabs - Account setup, first generation, and platform overview
• ElevenLabs Studio Tutorial - Create long-form audio projects with multiple speakers
• ElevenLabs - Full review with pricing breakdown and feature analysis

External Resources

• ElevenLabs Pronunciation Dictionaries Docs - Official reference covering API endpoints and dictionary file formats
• W3C Pronunciation Lexicon Specification - The PLS XML standard ElevenLabs implements
• IPA for English (Wikipedia) - Phoneme reference chart for writing accurate phoneme rules