Turn a script into subtitles that match your recording

A script needs a recording before it can have reliable timing

To turn a script into subtitles, first make the text match the final recording, then assign and review the start and end time of each cue. A plain-text script contains words but usually no information about pauses, delivery speed or edits. Renaming a .txt file to .srt cannot supply that information.

There are three practical routes: generate timed text from the recording and use the script to correct it; align a matching transcript with a service that supports that task; or create and time cues manually in a suitable caption editor. Choose the route before preparing the file. This guide is for scripts and recordings you own or have permission to process, adapt and deliver.

Check what you actually have

Keep the final media file, the script revision and the spoken language together. A shooting script may contain scene headings, camera instructions, deleted dialogue and words added during recording. A transcript describes what is actually audible. For alignment, that difference matters more than whether the document is beautifully formatted.

If no recording exists yet, keep the script as a production reference. You can plan short phrases, but any proposed timestamps remain provisional until the final take and edit exist. Do not divide the planned running time evenly across paragraphs and call the result synchronized subtitles.

If you already have a valid SRT or WebVTT paired with this edit, start from that timed file instead. Recreating its timing from a script would introduce unnecessary work. Use the separate timing-repair guide if an existing track is consistently early or late.

Choose how the timing will be created

For a local Polyglot workflow, import the finished audio or video and generate a timed subtitle draft from its speech. Keep the script beside you as a spelling and terminology reference while reviewing the result. Polyglot does not automatically align an untimed script or import a plain-text document as an already timed subtitle track. Its standalone subtitle imports are SRT and WebVTT.

If you need to keep the exact prepared transcript as the starting text, use a tool that explicitly supports transcript-to-audio alignment, or author the cues manually. YouTube documents an Auto-sync route for a transcript in the same language as the spoken audio, using a supported speech-recognition language. It advises against this approach for poor audio or videos over an hour. This is a YouTube upload and processing workflow, separate from local Polyglot processing. Its documentation also says the synchronized transcript is published automatically when ready; use it only when publication through that account is intended.

For manual timing, choose a caption editor that can create and split cues while you listen. Polyglot can edit text and adjust the timing of existing cues, but those controls do not turn an untimed paragraph into a sequence of new cues. Choose an appropriate authoring tool where the task requires that structure.

Prepare the words that belong to the recording

Work on a copy of the script. Remove production instructions from the spoken text, add audible departures and check that passages are in the order used in the final edit. Keep uncertain words in a separate review note until resolved; do not let a plausible script line replace unclear evidence from the recording.

For YouTube transcript input, its official instructions call for a plain-text file and recommend UTF-8 for non-English text. A blank line can request a new caption. Those are destination-specific preparation rules, not an SRT syntax recipe and not a promise that every aligner treats paragraph breaks the same way.

Relevant sounds and speaker information also need review. W3C describes captions as conveying the speech and non-speech audio needed to understand the content. An unspoken instruction such as “close-up of folded paper” does not become a sound caption. A meaningful audible event absent from the script may need a separately authored cue. Use the sound-and-speaker guide for deciding how to write it.

Example: a pause defeats evenly spaced timestamps

Imagine an original fictional folding demonstration. The script says “Fold the paper in half. Open it again.” The speaker says the first sentence, pauses while folding, and speaks the second sentence later. The example below assigns invented times solely to explain the distinction; it is not an app test or a measured recording.

The two sentences have similar lengths, but that does not mean they should occupy adjacent equal intervals. A converter that spreads the text over eight seconds could display “Open it again” while the speaker is still silently folding. A file can therefore have valid SRT structure and still give the viewer the wrong instruction at the wrong moment.

Use playback to establish when each phrase is heard, then review its display duration and the gap. The illustrative three-second first cue and two-second second cue are not universal duration rules. Keep phrase boundaries readable and check at normal speed. If the finished recording uses different words or a different pause, its timings and text must change.

Fictional script:
Fold the paper in half.
Open it again.

Illustrative timed output:
1
00:00:01,000 --> 00:00:04,000
Fold the paper in half.

2
00:00:07,000 --> 00:00:09,000
Open it again.

Review timing locally, not just at the beginning

After transcription or alignment, watch the complete track with the exact media file. Start with the first spoken phrase, then inspect pauses, repeated sentences, retakes, long silent passages and the final line. These points make a misplaced block easier to detect, but they do not replace a full pass.

If one phrase is wrong, inspect that phrase and its neighbors. A uniform time shift moves every cue by the same amount; it cannot repair one mistaken pause, restore omitted speech or align a reordered script. If a whole block belongs to a deleted take, return to the final recording and correct the correspondence before adjusting individual edges.

In Polyglot, existing cues can be moved or resized for the selected language. If the draft has joined several phrases into one unwieldy cue, or an interval needs a cue that does not exist, finish the split or new-cue authoring in a suitable editor. Never stretch an unrelated cue across missing speech merely to make the track look complete.

Translate the accepted track and inspect the delivered file

Approve the original words and timing before translating. A translated script is not a same-language transcript for an audio aligner. Polyglot can translate a reviewed timed track while preserving source timestamps during processing; every translation still needs a meaning and reading-length check. Manually adjusted language timing should be reviewed against the same recording.

Export the format the recipient accepts, then open that actual export with the final media. Check that all intended cues survived, that punctuation and characters display correctly, and that the right language track is selected. A subtitle file is separate from the video; Polyglot does not upload it or burn it into the picture.

Keep the source script, corrected transcript if used, final timed export and review note distinct. Record which route supplied timing and what remains unresolved. If the file plays correctly locally but has not been checked on the destination, leave destination acceptance pending. A script conversion is complete only when its text and timing have been checked against the intended recording.

Final review note
Media edit: _____________
Script revision: ________
Timing route: ___________
Audio language: _________
Full playback: __________
Unresolved passages: ____
Export file: ____________
Destination check: ______

Keep exploring.