Audio Proof Checker

The checker uses word alignment to narrow a long recording down to moments worth hearing. Missing words, added words, and replacements appear separately.

Speech recognition is not ground truth. Names, accents, music, and room noise can create false positives, so every issue can be played, selected in the script, or dismissed.

Before comparing, both texts are split into words, lowercased, and stripped of punctuation, so commas, capital letters and line breaks never raise a flag. The Word match percentage counts how many word edits separate the transcript from the script, relative to the script's length; treat it as a quick read on how much to listen to, not as a pass mark.

A practical order of work: fix substitutions first, because they are the mistakes listeners hear (a wrong name, a misread number). Then check missing words, which often mean the voice skipped a line after a long sentence. Unexpected words are usually transcription noise, but a run of them can mean the audio file belongs to a different version of the script.