Batch Instagram Transcript Review

Already have several Reel transcripts? Keep each source link paired with its raw text, then review the batch below. This local worksheet flags bracketed non-speech cues and immediately repeated words in up to 20 saved transcripts. With a trusted reference, it also counts word-level edits. It does not fetch links, transcribe videos, hear background music, or prove that plausible words are correct.

Review saved transcripts

Paste text you already have. A source link is optional but helps prevent two results from being mixed up. Nothing entered here is submitted to XAPN by this worksheet.

Start with one saved transcript. Up to 20 items; 5,000 characters per text field.

Three paired Reel sources and raw transcripts flow through visible-text checks into a per-item review log. Audio verification remains a separate human step.
A batch log keeps source, raw result, and warning count together. It is not a batch speech-recognition queue.

What the log can actually establish

For each nonempty item, the worksheet counts word tokens after ignoring letter case and punctuation. It flags only two visible patterns: bracketed or parenthesized non-speech cues such as [music], and immediately adjacent repeated words. These are review prompts, not errors by definition. The speaker might really repeat a word, and a music cue can be useful metadata.

When you supply a checked script or manual transcription as a reference, the worksheet reports the minimum word insertions, deletions, and substitutions needed to turn the draft into that reference. This is an edit count for this pair of texts, not an accuracy rate. An unverified second generator output is not a trusted reference. The CSV contains the source link, draft and reference word counts, visible warning count, edit count when available, and review notes. It intentionally does not include the transcript bodies.

Privacy and capacity: checks and CSV creation happen in this browser tab. The ordinary site analytics script still loads, but this worksheet makes no request containing your pasted text. Reloading loses the work; download the CSV before leaving. The 20-item and 5,000-character limits prevent a large comparison from freezing the page. Keep your original transcript files separately, since the CSV is only a review index.

Where a clean-looking row can still be wrong

Our earlier controlled music test provides a concrete counterexample. In one synthetic English sentence, run through one Whisper tiny English model at three music levels, the -35 dB speech-to-music sample changed the final word from “Friday” to “writing.” A structural scan of text alone cannot infer that substitution; it takes the audio or a trusted reference. This page does not turn that one sample into a general Reel threshold. At -40 dB in the same limited setup, the mismatch was larger, but neither number ranks outside services.

For a batch, the dangerous shortcut is to equate zero visible warnings with a checked transcript. Use the log to prioritize human review, then replay names, numbers, calls to action, and speech masked by music. A source link is a traceability aid, not proof that the link is public, reachable, or owned by the person pasting it.

What this is not

XAPN's Reel transcription path remains one link per run. This page reviews texts you bring to it; it does not queue multiple URLs or retrieve audio from YouTube or TikTok. The platform link boundary is unchanged. It also performs no vocal source separation: a background-music problem requires listening, a trusted reference, or a separately validated audio-processing step. We have not submitted a matched batch to SpeakApp, GetTheScript, or Eden, and this page does not compare or rank their accuracy or limits.

For a single item with a full explanation of the edit metric, use the Instagram Transcript Checker. For many saved items, use the worksheet above to keep every source and review decision aligned before you edit the original text.