ToolNavio
ENRO
Home/Guides/Guide
The script workshop · 4 MIN READ

Prepare Romanian text before generating a voice

Diacritics, proper names, numbers and pauses: a practical checklist for fewer revisions after generation.

Text that reads well on screen does not always work well aloud. Check the script before changing a voice’s settings repeatedly. This is a proposed editorial process; the examples are original and are not results of audio tests.

Keep a version specifically for speech

Work on a copy of the publication text. Expand abbreviations or write some numbers as words in the audio script without changing the article or subtitles. Keep the versions connected so that a content correction reaches both.

Add Romanian diacritics before testing. Look for ambiguous words and pronouns with unclear referents. A voice can pronounce a sentence perfectly even when the listener cannot understand it. Read the script aloud and mark where you need a breath or a brief explanation.

Rewrite numbers according to their meaning

In Romanian notation, “12.500”, “12,5” and “12:50” can mean very different things. In a trial script, replace them with the intended spoken form: “douăsprezece mii cinci sute” (twelve thousand five hundred), “doisprezece virgulă cinci” (twelve point five), or “ora douăsprezece și cincizeci” (12:50). Check grammatical agreement when a unit follows.

For dates, decide whether listeners need the full date or a shorter reference. Name the currency when it matters. Avoid a mechanical replacement of every number: a telephone number, software version and quantity are read differently.

Make a list of names and specialist terms

List people, places, brands and foreign words. Do not permanently alter a name’s spelling because one sample sounds wrong. Keep the correct text and try a pronunciation hint in the speech copy or the provider’s pronunciation controls.

Murf documents pronunciation options in Studio. Controls vary by product, voice and model, so consult documentation before inserting symbols or instructions. A command recognised by one engine might be read aloud by another.

Use sections you can revise

Split a script at complete ideas, rather than at arbitrary character counts. Short sections with clear beginnings and endings can be easier to correct, but very short fragments may lose the context needed for intonation. Try two section lengths and compare how they join.

Punctuation should support meaning. Do not add many commas just to slow a voice down. Simplify the sentence first, then use documented pause or speed controls. Save the settings so you can reproduce the chosen result.

Finish with three review passes

First, listen while following the script and note omissions, pronunciations and numbers. Second, listen without the text: the message should still be clear. Third, listen in the final edit with music or visuals if relevant. Each pass checks something different.

Correct one issue at a time and use clear file names, such as “lesson-01-script-v2-voice-a”. Avoid regenerating the whole project when the product supports a local correction. Ask for a language review when you cannot judge the result yourself.

  • The final script matches the information you intend to publish.
  • Names, dates and amounts have each been checked by listening.
  • No technical instructions are accidentally spoken.
  • The exported version is the approved version.

Sources and updates

Documentation checked on October 29, 2026. Providers may change features, pricing and terms.

Spotted information that has changed? Send a correction.

Software mentioned