Finding speech missing from an automatic transcript
Audio with no corresponding transcript text is flagged for review, with support for adding words and adjusting utterance boundaries.
Work done August 27 – September 1, 2026 · Last edited September 25, 2026
On this page
An automatic transcript can leave out speech. Arc flags stretches of audio that may contain speech without corresponding words in the transcript, drawing attention to places that need listening and transcription.
Reviewing missing words in context
Flagged stretches appear alongside the transcribed words, where they can be played and filled in after listening. They remain part of the sample when work is saved, so unfinished transcription is available when the sample is reopened.
Dividing utterances before every word is clear
A change of turn can be audible before the words themselves are clear. Arc now supports starting a new utterance at a flagged stretch without requiring the missing words to be transcribed first. This allows the transcript’s structure to be adjusted while the wording is still being worked out.
More precise playback and timing
Playback for an utterance stops at its boundary, and word playback uses that word’s assigned audio span. Adjusting a word boundary also updates an adjoining word’s boundary when needed, keeping the two spans from overlapping.
