Treat Shotcut’s speech-to-text output as a draft, not a finished caption track. Compare each subtitle with the original audio, correct the wording, check its start and end separately, and preview the corrected result before deciding whether to burn it in. This order matters because changing a subtitle’s appearance does not fix a mistimed cue, and burned-in text becomes part of the video image.
Review wording, then check each boundary
Shotcut’s Speech To Text uses AI based on OpenAI’s Whisper. The documentation warns that recognition can make occasional errors and that subtitle items sometimes start earlier than expected; it says cue timing comes from the model and tool. Results appear as a new top-level subtitle track in the Subtitles panel. Shotcut’s Speech To Text documentation
Use this ordered review:
- After the speech-to-text jobs finish, select the generated track in the Subtitles panel. Play the source audio and follow the cues in order. Check names, technical terms, numbers, negation, and any wording that sounds uncertain. Correct the subtitle text against what you can actually hear; do not infer unclear words from context alone.
- For each cue, compare its start with the first audible word and its end with the last. If the text appears before speech, adjust that cue’s start. If it remains after the line ends or cuts off the last word, adjust its end. Keep the two boundaries as separate checks.
- Play through the edited cues in context, including the neighboring cue before and after each correction. The Subtitles panel highlights the cue at the playhead, which can help locate the item being checked. Listen for a caption appearing too early, disappearing too soon, or overlapping a neighboring line.
- Add the Subtitle: Burn In filter to Timeline Output only if permanent on-image captions are required. The filter can preview the selected subtitle track with its styling. Inspect that preview, then export a short representative section and review the resulting file. A burned-in caption cannot be switched off by the viewer. If captions should remain editable or optional, consider exporting a separate SRT instead; it contains text and timing, not visual formatting. Shotcut’s subtitle workflow and export documentation
A cue-by-cue acceptance worksheet
The following is a hypothetical review record, not a claim about a completed Shotcut test or a guaranteed recognition result. The times and words are sample observations for deciding what to check next. Use the original audio and your project’s actual cue values.
| Cue and audio observation | Recognized text | Corrected text | Start–end check | Playback confirmation | Burn-in decision |
|---|---|---|---|---|---|
| A speaker says a sample name, “Mira Chen”; first sound is at 00:12.40 | “Mirror Chen” | “Mira Chen” | Hypothetical cue starts 00:12.10; compare with first sound | Replay with the previous cue; confirm the name is audible before it appears | Include only after text and timing pass |
| A sentence ends with the sample word “today” near 00:18.00 | “We can ship it” | “We can ship it today” | Hypothetical cue ends at 00:17.70; check whether the final word is cut off | Replay through the next cue; confirm the ending and transition | Hold if the ending is still clipped |
| A pause separates two spoken lines | Two cues appear close together | Keep wording only if supported by audio | Check each cue against its own speech, not a shared visual style | Listen across both cues for early display or overlap | Decide after the gap and both boundaries are reviewed |
A timing difference in this worksheet is a prompt to listen, not a universal tolerance. Record actual observations rather than assuming every cue needs the same offset.
Limits and what to do when a cue fails
Recognition is not proof that a word was spoken. If a name or phrase remains unclear after replaying the relevant audio, mark it for a human decision rather than silently guessing. Shotcut’s documentation does not promise that changing the model or language setting will correct a particular cue, so do not treat a rerun as verification.
If several cues begin early, inspect their individual starts before changing font, position, or other styling. If only one boundary is wrong, adjust that cue rather than shifting the whole track. If the text is correct but hard to read against the picture, address presentation separately in the burn-in preview; the filter’s styling controls affect appearance, not whether the words match the audio. If a cue is missing or its timing cannot be established from the recording, return to the audio and create or edit the subtitle item manually in the Subtitles panel.
The suggested short-section export is a review step, not a claim that any export has been tested here. Check the actual exported file: captions intended to be burned in should appear at the reviewed moments, and the lines should remain readable through the transitions. Do not make a full delivery decision from a styled preview alone.
Common questions
Should I fix a recurring name with Find and Replace? It can help when the same misspelling recurs, and Shotcut’s subtitle documentation describes search and replace as an option for repeated text errors. Still check the result against the audio; a global replacement may be wrong where the spelling or speaker differs.
Do subtitles show in the Player before I add Burn In? The documentation says subtitles are not displayed in the Player unless the Subtitle: Burn In filter is added. You can still review and edit cue text and timing in the Subtitles panel; use the filter when you need to preview the rendered appearance.
Must I burn captions into the video? No. Choose burn-in when captions must always appear in the image. If viewers need optional captions or the text may need later edits, keep or export a separate subtitle file when that suits the delivery.
Sources and next step
- Subtitles > Speech To Text - Documentation - English - Shotcut Forum
- View > Subtitles - Documentation - English - Shotcut Forum
- Subtitle Burn-In Video Filter - Documentation - English - Shotcut Forum
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.