Build a Custom Pronunciation Dictionary in Speechify

A text-to-speech voice can read every word on the page and still get the important ones wrong. Names, acronyms, product names, and technical terms often need a pronunciation rule.

Speechify Studio lets you add those rules through its Pronunciation Library. You enter the word, add its phoneme representation, preview the result, and save it for use in your project. The exact menu placement can differ by Speechify product, account, or plan, so check your current Studio settings if you don’t see the option.

Key Takeaways

• Use the Pronunciation Library for terms that appear across a project.
• Enter the word and its pronunciation in IPA or the format Speechify requests.
• Use inline pronunciation controls for a one-off correction.
• Test each entry across voices, sentence positions, and languages.
• Keep a shared pronunciation record for names, brands, and technical vocabulary.

Understand Speechify’s Pronunciation Options

Speechify pronunciation settings are mainly useful in Speechify Studio, where you create AI voice-overs, dubbing projects, and other audio content. The consumer Speechify reader and the Studio editor don’t always expose the same controls.

Start by identifying where the error occurs. If the same word appears throughout a video or script, add it to the Pronunciation Library. If one line needs a different reading, use the pronunciation control inside that text block.

The library is not a replacement dictionary. It tells the voice how to say a selected written term. The source text can remain unchanged, which helps when your script needs to preserve the official spelling of a company, product, or person’s name.

Speechify describes pronunciation adjustments alongside pauses and other voice-over controls in its AI voice-over guide. Review the current interface before building a large dictionary. Buttons and project types can change.

A pronunciation entry usually contains two parts:

• The written term, such as Kubernetes, Nguyen, or SQL.
• The phoneme representation, which tells the speech engine which sounds to produce.

Speechify’s current Studio workflow uses IPA, the International Phonetic Alphabet, for phoneme input. IPA is more reliable than a casual spelling because English spelling rules vary. The word read, for example, has different pronunciations in “I read the report” and “I read the report yesterday.”

Add a Pronunciation Entry in Speechify Studio

Use this workflow for a term that appears more than once in a project.

1. Open a Studio project. From the Home Screen, select New Project. Choose AI Voice Over for text-to-speech work or Dubbing for a video project. If your account shows different labels, open the closest voice-over or dubbing editor.
2. Open the Pronunciation Library. Look in the toolbar or project settings for Pronunciation Library. The control can appear in a different location depending on the editor and account.
3. Create a new entry. Select Create New Pronunciation. A dialog should appear with fields for the written word and its phoneme representation.
4. Enter the term and pronunciation. Add the exact form that appears in your script. Then enter its IPA pronunciation in the format accepted by the field. For example, YouTube may use /ˈjuːtuːb/.
5. Preview the result. Select Preview and listen to the sample. Check the stress, vowel sounds, syllable breaks, and ending consonants.
6. Save the entry. Select Save after the preview sounds correct. Return to the project and play the relevant sentence again.

The written term needs to match the text that Speechify scans. Capitalization may matter in some editors. Punctuation, plural endings, and possessives can also affect how the engine applies an entry.

Use a pronunciation dictionary for recurring terms such as these:

• OpenAI, pronounced as the letters and word combination your audience expects.
• SQL, pronounced either “sequel” or “S-Q-L,” depending on your company standard.
• Kubernetes, commonly pronounced close to /ˌkuːbərˈnɛtɪs/.
• Xiaomi, which needs testing because English-language voices may use different approximations.
• Nguyen, which varies by speaker, region, and personal preference. Confirm the person’s preferred pronunciation before recording.

IPA symbols can be difficult to enter on some keyboards. Copy the pronunciation from a reliable dictionary or phonetic reference, then listen to the result. Don’t assume that a technically correct IPA form matches your chosen voice’s accent.

You can also correct one word without creating a reusable library entry. In the Voiceover Studio text block, highlight the word, choose Pronunciation in the right-side toolbar, enter the IPA form, and press Enter. Use the play button beside the text box to test the change. A Speechify Studio beginner tutorial can help you locate the main editing controls if your workspace looks unfamiliar.

Test the Dictionary Across Voices and Contexts

A pronunciation entry isn’t finished when one preview sounds acceptable. Speechify voices use different accents, pacing, and vocal patterns. A term that works in one voice may sound too compressed or unnatural in another.

Test the entry with the voice you plan to publish. Then test it with at least one other relevant voice. A US English voice, a UK English voice, and a multilingual voice may handle the same phoneme sequence differently.

Use three short test sentences:

1. Place the term at the start of a sentence.
2. Place it in the middle of a sentence with commas or parentheses.
3. Place it next to a number, acronym, possessive, or punctuation mark.

For SQL, test “SQL improves reporting speed,” “Our SQL database stores customer records,” and “The SQL team’s dashboard is ready.” These contexts reveal whether the saved rule applies to the exact word form you need.

Test names with the surrounding words that will appear in production. “Dr. Nguyen joined the call” may sound different from “Nguyen’s account is active.” A name in a list may also receive different stress than the same name in a full sentence.

Multilingual content needs another test. A French, Spanish, Japanese, or Arabic term may need a voice that supports the relevant language. Changing the phoneme entry won’t fix an unsuitable voice model. Select a voice with the right language and accent first, then adjust the term.

A pronunciation rule is only useful when the final voice says the word correctly inside the full sentence.

Export a short sample before rendering a complete video or audiobook chapter. Listen through headphones and speakers. Check words near music, transitions, and fast narration. This catches errors that a short isolated preview can miss.

For teams that connect Speechify to another voice workflow, LiveKit’s Speechify TTS guide documents custom pronunciation through SSML. That approach applies to an integration workflow, not necessarily to the Studio interface. Don’t assume an SSML rule will automatically appear in your Speechify project library.

Choose Between Library Entries and Inline Corrections

Use the saved library when consistency matters. Use inline pronunciation when the correction belongs to one passage.

A recurring product name should have one approved entry. A one-time guest name may only need an inline adjustment. This distinction prevents a local exception from changing every use of the word in a project.

Inline pronunciation is also useful when the same spelling needs two readings. The acronym API might be spoken as “A-P-I” in a developer training video and as a company-specific term in another project. A library-wide rule could create the wrong result in one of those contexts.

Keep the official spelling in the script. Don’t replace Kubernetes with “koo-ber-net-eez” unless Speechify’s controls fail and you need a temporary workaround. Replacing text can affect captions, search, screen-reader output, and translated versions.

For business use, store your approved terms outside Speechify as well. A spreadsheet or internal document can include:

• Written term
• Approved pronunciation
• IPA representation
• Preferred language and accent
• Owner or source of approval
• Date last tested
• Projects that use the entry

This record helps editors apply the same rule across separate projects. It also gives a new team member a clear starting point instead of forcing them to guess from an audio preview.

Speechify’s pronunciation video maker is another relevant Studio workflow for pronunciation-focused content. Treat its controls as project-specific until you confirm that a saved entry is available in your other Studio projects.

Fix Common Pronunciation Problems

If Speechify ignores an entry, check the written match first. The library may contain OpenAI while the script uses openai, OpenAI's, or OpenAI-powered. Create the required variants only after testing the original rule.

If the voice produces the wrong stress, adjust the IPA representation. Stress marks can change the result. Listen to the complete word, not only its first syllable.

If the pronunciation sounds correct in Preview but fails in the script, inspect the text block. The term may contain a hidden space, a different apostrophe, a line break, or a punctuation mark that changes how Speechify reads it.

A language mismatch is another common cause. English phonemes won’t always produce a natural result in a Spanish or French voice. Select the intended language voice and test the term again.

Some pronunciation controls may be limited to Speechify Studio, certain project types, or particular plans. The feature may also move as the product changes. Check the Pronunciation Library in your current workspace and verify availability in Speechify’s documentation or account settings before designing a production process around it.

When a library rule still fails, use an inline correction for the affected text block. If the issue remains, split the sentence, adjust punctuation, or test another voice. Keep the final audio sample with your project review notes.

Conclusion

A custom Speechify pronunciation dictionary gives your scripts a repeatable way to handle names, acronyms, brands, and technical terms. Add recurring corrections to the Pronunciation Library, use inline controls for local exceptions, and keep the original spelling in your source text.

The final check is always audio. Listen across the voices and contexts your audience will hear. A saved pronunciation entry is ready for production only when the complete sentence sounds right.