Two common starting points
A subtitle translator usually starts with one of two kinds of source material, or sometimes both. One is a written script: the lines as they were planned or recorded. The other is the audio itself, meaning the spoken words as they appear in the finished video. These look similar but behave differently, and the choice shapes how the work goes.
What a script offers
A script is text, so it can be read quickly, searched, copied into a translation tool and checked for spelling. It usually names the speakers, shows how names and technical terms are written, and sometimes includes stage directions, context notes or descriptions of sound.
For a translator, this is a real advantage. There is no need to guess the spelling of a name or decide what a muffled word was. The translator can read the whole text beforehand, understand the structure and plan consistent terminology.
Where scripts mislead
A script is a plan, and finished videos often depart from the plan. Common differences include:
- Speakers who improvise, skip lines, add remarks or change wording.
- Lines cut or reordered in editing.
- Late changes made after the script was last updated, so the version on file is older than the final cut.
- Interviews and live sections that were never scripted.
- Scripts that describe what is meant rather than what is said, with no hesitations, repetitions or interruptions.
If subtitles are translated from a script that no longer matches the audio, viewers can see words on screen that nobody speaks, or hear words that are missing from the text. Even small differences are noticeable to people who hear the language, and bilingual viewers may lose trust in the subtitles. For this reason, it is sensible to compare script and audio, rather than assume they agree. The topic of how transcripts differ from scripts and subtitles is worth reading on its own.
What translating from the audio offers
Working from the audio means working from what the viewers will actually hear. It captures changes, ad-libs, interruptions and the way people really talk. It also lets the translator catch tone, pace, emphasis, sarcasm and emotion, which a flat script cannot show.
The video also provides visual context. A character might point at something, or a picture on screen might clarify an ambiguous word.
Where audio makes things harder
Audio is demanding. Several problems come up again and again:
- Audio quality. Background noise, music, overlapping speakers and low volume make words hard to catch. Poor audio is a common cause of mishearing, and the same issues limit automatic speech recognition.
- Accents and regional dialects. Unfamiliar pronunciation may be difficult to follow without help.
- Names and technical terms. Heard but not seen, a name may be spelled in several plausible ways. Terms from specialized fields may be misheard as ordinary words.
- Unclear speech. Mumbling, fast talk and unfinished sentences force the translator to infer meaning.
- Time. Listening, replaying and checking every line takes longer than reading text.
If a transcript is made from the audio before translation, any errors in it flow into the translated subtitles. Transcribing carefully, or having the transcript checked, is therefore an investment.
A combined approach
In practice, many projects use both. A frequent pattern is:
- Start with the script for names, terms and structure.
- Listen to the audio to confirm the actual wording and note differences.
- Use the audio as the final authority for what is said.
- Mark places where the two disagree, and ask the client or producer when it matters.
- Keep a list of names and terms so the same spelling is used throughout.
This works well because each source covers the other’s weakness. The script gives certainty on spelling; the audio gives certainty on content.
Special cases
- Dubbed or re-voiced material. The script may relate to a previous version of the audio. Check which version the subtitles must match.
- Songs and foreign words. If characters sing or switch languages, decide in advance whether these lines should be subtitled. Untranslated speech can be a deliberate part of the scene.
- Live or unscripted events. Translating from a recording may be the only option, which calls for time and careful listening.
- Material with technical or sensitive content. For medical, legal or educational video, an inaccurate source leads to inaccurate subtitles. Important content deserves a careful transcript and review by a qualified human reviewer.
Questions to ask before starting
- Is there a script, and does it match the final cut?
- Which version of the video are the subtitles for?
- Are there lists of names, terms or previous translations?
- Who can answer a question about an unclear line?
- Are there on-screen texts that need translation too?
What the choice means for quality
Neither starting point is automatically better. A reliable, up-to-date script can speed up the work and reduce spelling mistakes. A faithful treatment of the audio can produce subtitles that match what viewers hear. What goes wrong is when no one checks. Treating the script as final without listening, or working only from poor audio without checking available text, are both avoidable errors. A short comparison pass before translation begins often costs far less than fixing mismatches afterward.