A practical guide
Writing for ElevenLabs: how to set pauses and check pronunciation
A reader can quickly understand “15.05, 18:30, 2500 ₽”. A narrator needs text with an unambiguous spoken form. Here is how to prepare a short voiceover, from writing numbers out to pauses and checking the finished file.
Prepare a separate version of the text for listening
Do not send a video caption straight to speech generation without rereading it. On screen, numbers, brackets and line breaks help the reader. In audio, the listener hears one sequence of words and may not have time to reconstruct their connections.
Example line: “MC 15.05 at 18:30, 2500 ₽, materials incl.” Spoken version: “The masterclass takes place on the fifteenth of May. It starts at eighteen thirty. Participation costs two thousand five hundred rubles. Materials are included.” This is a fictional announcement, not a service offer. We have expanded abbreviations and made each number’s meaning explicit.
Keep both versions: on-screen text and the approved narration script. If the script’s price or time changes, update related captions immediately. Otherwise, even a well-narrated video will give conflicting information.
Test a difficult passage first
Choose a voice and a model, then prepare a few sentences containing the trickiest material: a surname, a proper name, an abbreviation and a number with its unit of measurement. Listen to them before generating the whole script. For an abbreviation, decide in advance how it should sound to the listener. API can be spoken letter by letter or replaced with the words “software interface” if that suits the audience. To ensure an accurate reading, write the amount “2,5 million” as “two and a half million” in the script. Check the meaning and grammatical case in the complete sentence.
Test the challenging passages that will appear in the final recording.
Pauses depend on the selected model
Start with ordinary punctuation. End a complete thought with a full stop and separate a short list with commas. An ellipsis can make a line sound hesitant, so do not scatter it everywhere simply to create silence. Make the sentence clear first, then adjust the delivery.
ElevenLabs has no universal pause markup for every model. Its documentation states that Eleven v3 and Eleven v4 do not support SSML break. The break tag is described for other compatible modes, but overuse can cause instability. Do not copy it into any selected voice and expect an identical effect.
If a pause needs to line up precisely with a video edit, use the Studio timeline: you can separate phrases and adjust the gap there. An exact silent interval on the timeline and an expressive pause in speech serve different purposes. Check both in the finished clip.
Create a pronunciation rule for a recurring name
Studio includes Pronunciations Editor and attachable dictionaries. An alias rule substitutes a word or phrase for pronunciation. When an unusual name appears throughout a project, keeping the correction in one place is easier than changing every occurrence manually.
Phonetic-markup support differs between models and tools. Check the chosen mode’s documentation, then listen to the word in several phrases. Even compatible transcription still requires checking stress, word endings and the particular voice’s intonation.
Evaluate the result by listening and checking it against the script
Listen to the entire audio while following the text: check for missing words, correctly spoken numbers and pauses that change meaning. After regenerating one section, check transitions to adjacent phrases. A sudden pace change may be more noticeable than one misplaced word stress.
Keep the approved script, model, voice and pronunciation rules beside the final file. At the next edit, you will know which text is actually spoken and which section needs updating.
