When the Subtitles Do Not Match the Audio: Repair Your Shadowing Transcript
Subtitles don't match the audio? Classify the mismatch, patch only the verified words, separate reductions from text, then re-shadow the repaired line.
For shadowing, follow the audio you are actually imitating—but classify the mismatch before changing the transcript.
The subtitle says “I will call you later.” The speaker says “I’ll give you a call later.” Neither sentence is bad English, but only one predicts the words you need to shadow in that audio.
First: subtitles are not always word-for-word transcripts
A subtitle can be professionally made and still be a poor word-for-word shadowing script. Current BBC Learning English transcript PDFs can explicitly say that they are not a word-for-word transcript. Netflix’s current subtitle-template guidance likewise says templates are not expected to be verbatim and may be edited for reading speed through deletion, reformulation, re-timing, re-segmentation, merging and other techniques. See a current BBC Learning English transcript example and Netflix’s subtitle-template guidance.
So the first question is not, “Which one is wrong?” It is, “What kind of text am I looking at?”
Label 1: the subtitle paraphrases or condenses the audio
This is the easiest mismatch to misdiagnose. The subtitle communicates the right idea, but the speaker uses different words. For ordinary viewing, that may be completely acceptable. For exact shadowing, the learner needs the sequence that is actually spoken.
Original practice example
I'll give you a call later.
A subtitle might say “I will call you later.” Both sentences are grammatical and express a similar intention, but they are not the same wording. Give someone a call means call someone, and I'll is the contraction of I will. If the audio says “I'll give you a call later,” that is the wording your small shadowing transcript should use.
Use give someone a call as a natural way to say you will phone someone.
Listen without the subtitle, confirm the spoken phrase, write only that short line in your practice copy, then shadow it once.
Explore more language-learning guides in Media-Based Language Learning.
Notice what you did not do: you did not declare the original subtitle bad. You simply made a more useful practice transcript for a different job.
Label 2: several lines differ—check the track or version before editing
If one line differs, patching one line may be sensible. If many nearby lines keep using substantially different wording, stop editing and check the setup first.
On dubbed content, subtitles and audio may come from different workflows. Netflix’s U.S. English SDH guidance says English dubbed audio or its dubbing script should be the basis for matching SDH, while still allowing limited reduction when reading speed or synchronisation requires it. External subtitle files can also belong to a different release or edition; Plex’s official guidance warns that subtitle files need to match the particular version of the media. See Netflix’s U.S. English timed-text guidance and Plex’s local-subtitle guidance.
Before repairing ten lines, check:
- the selected audio language;
- the subtitle or caption type;
- whether you are listening to a dub or the original audio;
- and, for local media, whether the subtitle file matches that exact edition.
Repair the practice copy, not the universe.
Label 3: the automatic caption may genuinely be wrong
Automatic captions are another case entirely. YouTube’s current help says machine-generated captions can misrepresent speech because of factors such as mispronunciations, accents, dialects or background noise, and it tells creators to review and edit incorrect transcriptions. See YouTube’s automatic-caption guidance.
That means one strange caption word is evidence to investigate—not evidence that your ears must surrender.
Original practice example
Could you send it by Friday?
By Friday gives a completion deadline. Until Friday is grammatical for duration, but it does not express the same one-time deadline. If an automatic caption drops by, replay the phrase and use context or reliable reference material before patching that word. If you still cannot verify it, keep the position marked as unclear rather than inventing a confident transcript.
Use this as a polite request with a deadline.
Hide the caption, listen again, then compare. Patch by only when you can verify that it is actually in the audio.
Label 4: the words match, but the timing does not
If the subtitle contains the right words but appears noticeably before or after the speech, you have a timing problem—not a wording problem.
Do not “fix” the sentence. For shadowing, use a better-synchronised track when available or make a small cue note for your own practice. Professional subtitle standards treat timing to audio as a separate concern from text accuracy; Netflix’s timing guidance, for example, explicitly requires subtitles to be synchronised with the audio. See Netflix’s subtitle-timing guidance.
The useful question is: Do the words need repair, or only the moment when I see them?
Label 5: connected speech only sounds like missing text
This is where learners can accidentally turn a good transcript into phonetic spaghetti.
Natural English connects and reduces words. A weak word may become very short. Sounds may link across word boundaries. None of that automatically means the lexical word disappeared.
Original practice example
I didn't mean to interrupt.
Mean to + verb expresses intention. “I didn't mean interrupt” is wrong in standard English because to is required. In fast speech, to may be weak and tightly connected to the surrounding words. Keep the lexical transcript as I didn't mean to interrupt. If useful, add a separate pronunciation note saying that mean to is compressed or linked.
Use this to apologise for an unintended interruption.
Shadow the correct lexical sentence. Do not rewrite the transcript into invented spellings simply to imitate reduction.
British Council shadowing guidance explicitly directs attention to pronunciation and connected speech, and research on text-presented shadowing recognises that a written script can help learners keep up with difficult audio. The text layer and the pronunciation layer can support each other without becoming the same thing. See the British Council shadowing activity and Hamada and Suzuki’s review of shadowing techniques.
What kind of mismatch are you hearing?
Classify it before you edit it
The subtitle and audio express the same broad idea but use different words.
PARAPHRASE OR CONDENSATION. The subtitle may be serving reading or translation rather than exact transcription. For shadowing, patch only your small practice copy to the verified spoken wording.
Several nearby lines keep using substantially different wording.
CHECK TRACK OR VERSION. Verify audio language, subtitle type, dub, media edition or local subtitle file before editing line by line.
One caption word is bizarre or does not fit the sentence.
POSSIBLE AUTOMATIC-CAPTION ERROR. Replay without text, use context and reliable reference material, and correct only what you can verify.
The wording matches, but the subtitle appears early or late.
TIMING. Keep the wording. Adjust the practice cue or use a better-synchronised track rather than rewriting the sentence.
The words are grammatical, but the speaker seems to swallow one of them.
CONNECTED SPEECH OR REDUCTION. Keep the lexical word and, if needed, put the reduction in a separate pronunciation note.
I still cannot tell what one word is.
UNRESOLVED. Mark that position as [unclear] in your private practice note and move on. Your ears are not autocomplete.
Patch only the words you can verify
Your shadowing transcript does not need to become a publication-quality subtitle file. It needs to be a small, trustworthy support for the line you are practising.
A useful patch has two layers:
- Lexical text: the actual words you can verify.
- Pronunciation note: optional reminders about reduction, linking, stress or timing.
Keep those layers separate. If one word remains genuinely unclear, leave it marked as uncertain and practise the verified material around it. A visible uncertainty is better than a confident invented word.
Original practice example
You don't have to decide right now.
Don't have to means there is no necessity. Mustn't is also grammatical, but it usually means prohibition. If a different subtitle track says You mustn't decide right now while the audio you are shadowing clearly says You don't have to decide right now, those are not interchangeable versions: the meaning changes. Patch your practice transcript to the verified audio, then check whether the broader track/version is mismatched before fixing more lines.
Use don't have to to reassure someone that an immediate action is unnecessary.
Shadow the verified audio wording once. Then hide the transcript and confirm that you can still hear the phrase you wrote.
Prove the repair against the audio
A patched transcript is not finished because it looks plausible. It is finished when it helps you follow the actual recording.
- Shadow the repaired line once with the text visible.
- Replay the same line with the text hidden.
- Check whether the words you wrote now predict what you hear.
- If the mismatch remains, return to the label step instead of inventing a new word.
Repair the shadowing transcript, not every subtitle
For shadowing, the audio is the performance you are imitating. That makes the audio the reference for the small practice transcript—not proof that every subtitle difference is an error.
Hear. Label. Patch. Prove. Keep the actual words separate from notes about how those words sound, and keep uncertainty visible until you can resolve it.
That gives you something much more useful than a “perfect” subtitle: a transcript you can trust for the next shadowing pass.
Sources
- BBC Learning English, current transcript example
- Netflix, “Timed Text Style Guide: Subtitle Templates”
- Netflix, “English (USA) Timed Text Style Guide”
- Netflix, “Timed Text Style Guide: Subtitle Timing Guidelines”
- YouTube Help, “Use automatic captioning”
- Plex, “Adding Local Subtitles to Your Media”
- British Council, “Shadowing”
- Hamada and Suzuki, “Situating Shadowing in the Framework of Deliberate Practice”