AFTER TRANSCRIPTION · REVIEW GUIDE
Check an Instagram transcript
before you quote it.
Replay the passage and check negatives, numbers with their units, names, speaker turns, and the sentence around the quote. Keep the raw transcript unchanged. Publish the wording you checked against the audio, not the version that merely reads best.
Already have the words? This page helps you review them. To generate a new transcript from a public Instagram video, use the tool.
A review pass that leaves a record
- Save the raw result. Copy it into a separate document before correcting anything. Give your checked version a different name so the first output remains available.
- Find the exact passage. Reopen the original public video and listen before and after the sentence. Note the playback position yourself; this site’s TXT output does not supply verified timecodes.
- Check the words that change the claim. Start with a negative, amount, unit, name, speaker change, or condition. If a word remains unclear, mark it unresolved instead of filling the gap.
- Separate quote from rewrite. A summary can compress a passage, but it should not be presented as a word-for-word quotation. Keep any visual annotation separate from the spoken text.
- Record the source and decision. Save the public URL, passage location, raw text, checked wording, and remaining uncertainty in the blank review record.
Eight edits worth catching
Every sentence pair below was written for this guide. The draft errors were inserted deliberately; they were not produced by an Instagram video or a speech-recognition model. These are review exercises, not an accuracy benchmark.
CASE 01 · CONSTRUCTED EXAMPLE
A missing negative reverses permission
I do not publish client names.
I do publish client names.
Replay the verb together with the word immediately before it. The missing “not” reverses the statement; adding polished punctuation would not repair it. Keep the negative in your checked quotation and do not infer permission from the shorter draft.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: not
Draft-only tokens: (none)
This output locates text changes. It does not decide which version matches a real recording.
CASE 02 · CONSTRUCTED EXAMPLE
A similar sound changes the budget
Budget fifteen dollars for the draft.
Budget fifty dollars for the draft.
Listen specifically to the amount, then write it beside the currency. Here “fifteen” and “fifty” are different instructions. A readable sentence cannot decide which amount the speaker intended. Do not silently choose the more plausible budget.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: fifteen
Draft-only tokens: fifty
This output locates text changes. It does not decide which version matches a real recording.
CASE 03 · CONSTRUCTED EXAMPLE
The value stays, but the unit changes
Leave a gap of fifteen milliseconds.
Leave a gap of fifteen seconds.
Audit the quantity and its unit as one phrase. Both lines retain “fifteen,” so checking numbers alone misses the problem. Replay the full instruction before copying it into a tutorial, brief, or caption. Preserve the unit actually spoken.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: milliseconds
Draft-only tokens: seconds
This output locates text changes. It does not decide which version matches a real recording.
CASE 04 · CONSTRUCTED EXAMPLE
A plausible name still needs a source
Ask Mira Vale before posting.
Ask Meera Vale before posting.
Mira Vale is a fictional name written for this exercise. The similar-looking draft is not a verified alternative spelling. In a real clip, check the speaker’s own written identification when available; if you cannot establish the spelling, mark it unresolved instead of guessing.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: mira
Draft-only tokens: meera
This output locates text changes. It does not decide which version matches a real recording.
CASE 05 · CONSTRUCTED EXAMPLE
The speaker corrects the instruction
Send three drafts—sorry, two drafts.
Send three drafts.
Replay beyond the first complete sentence or number. The correction changes the requested delivery. Keep the correction in a verbatim transcript; if you write a summary, report the final instruction and label it as your summary rather than presenting the shorter draft as an exact quote.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: sorry two drafts
Draft-only tokens: (none)
This output locates text changes. It does not decide which version matches a real recording.
CASE 06 · CONSTRUCTED EXAMPLE
The words move to the wrong speaker
A: Can I publish it? B: Not yet.
A: Can I publish it? Not yet.
Compare speaker turns, not just the combined words. The draft assigns the reply to A. Listen through the change of voice and retain separate turns when attribution matters. A word comparison can flag a missing label, but it cannot identify a person from the audio.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: b
Draft-only tokens: (none)
This output locates text changes. It does not decide which version matches a real recording.
CASE 07 · CONSTRUCTED EXAMPLE
A visual note becomes invented speech
[On screen: Step 2] Spoken: Wait until tomorrow.
Step two. Wait until tomorrow.
The bracketed line is an author-written visual annotation, not part of the spoken sentence. Keep visible text in a separate note. For a real Reel, replay with sound before putting a phrase inside quotation marks; a readable overlay does not prove someone said it.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: on screen | 2 spoken
Draft-only tokens: two
This output locates text changes. It does not decide which version matches a real recording.
CASE 08 · CONSTRUCTED EXAMPLE
A short extract drops its condition
Use the clip only if the speaker has approved the edit.
Use the clip.
Replay the end of the sentence, including any “only if,” “unless,” or exception. This shortened draft removes a condition. Save the complete checked passage before writing a shorter quotation. If the audio ends before you can hear the condition, leave the claim unverified.
Actual word-comparison output
Wording difference: yes
Reference-only tokens: only if the speaker has approved the edit
Draft-only tokens: (none)
This output locates text changes. It does not decide which version matches a real recording.
Two controls: different formatting, same words
We also compared I won't post it.
with I won’t post it.
, and Check the name.
Then replay the line.
with the same sentence on one line. Both returned no wording difference under the chosen token rules. That result concerns the text only; neither version was checked against audio.
The comparison changes curved apostrophes to straight ones, ignores case, and splits text into word tokens. Punctuation and whitespace between tokens do not count. It can therefore miss punctuation-only changes that affect interpretation. Do not use its output to approve a quotation automatically.
What we actually ran
On October 31, 2026, we ran a Python word comparison on these 10 constructed pairs. It reported wording differences for the 8 deliberately altered drafts and no wording difference for the 2 formatting controls. Each result is included in the sample data. This is a reproducibility check of our examples, not a recognition score, an error frequency, or a test of another service.
Download the sentences and actual comparison results (JSON), the example table (CSV), and the standard-library Python script. Put the JSON beside the script and run python reproduce.py. Nothing is uploaded and the script makes no network requests.
Questions before you publish
Does a clean transcript prove the quote is correct?
No. In the constructed examples above, several altered drafts are readable sentences. Readability alone does not settle which words were spoken. Check against the original audio.
What if I cannot hear the word clearly?
Leave it marked as unclear. Keep the source location and the raw wording so you can revisit it. Do not label the passage checked while its important wording remains unresolved.
Can I use the TXT file as timed subtitles?
A plain TXT transcript does not establish caption timing. This tool downloads text; it does not certify speaker labels, timecodes, translations, or permission to reuse a clip.