Dramatic-reading subtitles. Follow the performed words.

Subtitle a performance, with the script beside it

A dramatic reading can follow a written text closely and still change a word, repeat a phrase or leave a long silence. Subtitles need to represent that recorded performance. Polyglot can generate timed text from authorized audio or video, translate it locally on an Apple Silicon Mac, and export separate SRT or WebVTT files after review. It does not align a supplied script automatically or decide which departures are intentional.

This workflow is for a recorded reading of your own writing or material cleared for your intended use. It also works as a review process when one reader moves between a narrator and several characters. Live stage captioning, literary translation from a book file and automatic character recognition are different tasks.

Clear the text, edition and recording before processing

Keep a private record of the text edition, any adaptation or translation used, performer permission and the authorized recording and distribution. Check that the intended subtitle translations and publication are covered. Owning the camera file alone is not the clearance record for everything performed in it. Include any music or borrowed illustrations in that review, and hold unresolved material before release.

If you are using a public-domain text, check the specific edition or translation and the places where you plan to distribute it. Project Gutenberg explains that its public-domain assessments follow US law and may differ elsewhere. A freely downloadable text is not blanket clearance for a modern translation, another person’s performance or a third-party recording. This page does not determine the rights in any particular work.

Make a performance-to-script decision log

Finish the chosen take, retakes and cuts in your media editor first. Name the final recording and the script revision used by the reader. Generate subtitles from that recording, keeping the script open as a reference for names and unusual words. A printed line can help resolve a misheard word, but it is not evidence that the performer actually said it.

In a separate document, record the time, script wording, audible wording and decision. Correct recognition errors to match the audio. Retain an intentional change, repetition or false start that belongs to the accepted performance. If the creator wants a performance mistake removed, resolve the recording before finalizing its subtitles; changing only the text leaves the mismatch audible.

This original fictional example distinguishes a missing word in the draft from an accepted change in the performance. The log is a manual review aid, not a Polyglot feature or a claimed test result.

Fictional reading review
Take: lantern-v03.mp4
Script: original revision 2

00:18
Script: “Leave the lamp.”
Heard:
“Leave the lamp. Please.”
Keep the added “Please.”

00:41  Heard: “I will not go.”
Draft: “I will go.”
Restore the audible “not”

01:06  Script: [door closes]
Heard: no door sound
Do not invent a sound cue

Distinguish the reader from the character

A change in voice may signal a character, quoted speech or a return to narration. Check those transitions against the recording and the creator’s interpretation. A cast list is useful for consistent spelling, but the audience may not yet know who is speaking. DCMP’s speaker guidance advises against introducing a name before it is established in the audio or on screen, and distinguishes a portrayed character from the performer.

Agree on a restrained identification convention for the destination. Where identification is needed, manually review the text labels and their timing. Do not label every line with the performer’s real name when the line belongs to a character. Polyglot does not automatically identify speakers or characters; a change in vocal pitch is not a reliable identity decision.

A silent gesture or an unspoken stage direction is not dialogue. Keep script annotations out of spoken subtitles unless they describe meaningful audio actually present. DCMP treats relevant sound effects as a separate captioning task. If an essential sound needs a new timed cue, finish it in an editor that supports individual cue creation. Visual description is a separate deliverable; transcript export does not supply it.

Keep pauses and translations faithful to the reading

Watch the end of a line through the next audible phrase. An early subtitle can reveal the answer to a dramatic question before the performer speaks it. A subtitle left on screen through a deliberate silence can also change how that silence reads. Review repetitions, interruptions and sentence endings in playback instead of mechanically copying the printed punctuation.

Polyglot preserves source timing during translation. Its uniform time shift can correct a constant offset; it does not reshape individual cues around a pause or split a long speech into a new rhythm. Use a suitable subtitle editor for individual retiming or restructuring, then verify the finished track against the same final take.

Give a fluent language reviewer the performance and the accepted deviation log. Preserve who addresses whom, repeated motifs, negation and the difference between a narrator’s statement and a character’s line. A familiar published translation may describe another edition or omit the performer’s added words. Do not paste it over the recording simply because it sounds more literary.

For verse or wordplay, have the creator and language reviewer decide what the audience needs to understand within the available time. Polyglot does not guarantee meter, rhyme or dramatic interpretation. Cue-structure checks do not approve the meaning or artistic choices.

Accept the subtitles in the player viewers will use

Export a reviewed file for each language, named with the final take identifier. Keep the timed originals, accepted deviation log and any separately edited reading transcript together. A script file and a performance transcript are different records; label them clearly so a later retake is not paired with old captions.

For YouTube, its documented subtitle workflow lets you select the video and language and upload a supported file with timing. Other destinations need their own format and display check. Polyglot exports sidecar files; it does not upload them, burn text into the video or create a theatre caption display.

Play the delivered video with each track selected. Check the opening and last line, every change between narration and character, all accepted script deviations and the passages with long pauses. Verify that names are not revealed early, repeated words remain present, and wrapping is readable at a narrow viewing size. Review meaningful sounds as well as speech: W3C explains that captions include audio information needed to understand the recording, so an unchecked speech transcript alone is not the finished accessibility review.

Media and subtitle processing stay local in Polyglot. Requested downloads, user-controlled updates, default-on content-free analytics with an opt-out and Pro licensing have separate network boundaries. New AI subtitle jobs use a seven-day trial and require Pro afterward; editing and exports remain free.

Keep exploring.