Transcript
Also known as: transcription, speech text
The written text of what was said in a recording, with timings, which is both the basis of captions and a searchable index of the footage.
Updated
What it means
A transcript is speech converted to text, normally with a timestamp against each word or phrase. The timings are the part that makes it useful for editing rather than merely for reading.
With them the audio becomes navigable. A phrase can be found by Semantic video search and the footage positioned at it, which turns a long interview or a piece to camera into something searchable instead of something to scrub through.
It is also the input to several other things rather than an end in itself. Captions, keyword search and subject detection all derive from it, which is why Speech-to-text usually runs once and feeds everything.
How it works in practice
Accuracy requirements differ by use. A rough transcript is fine for finding a moment or weighting a clip, while anything shown on screen through Subtitles and captions has to be correct because errors are visible.
Timings want checking at the start and end of long clips, since drift accumulates and a caption a second late is worse than no caption at all for anybody relying on it.
Keeping the transcript alongside the footage rather than only in the finished video pays off later, because it stays searchable long after nobody remembers which card held which conversation.
What to watch out for
Action footage transcribes badly and the reasons are physical. Wind, distance, helmets and the fact that people shout rather than speak all reduce accuracy well below what quiet recordings achieve.
Confident errors are the dangerous kind. A transcript that invents a plausible word reads as correct, and nothing in the output marks it as a guess rather than a recognition.
Names and jargon fail predictably too. Place names, trick names and brand names are exactly what a highlight wants to caption and exactly what general recognition gets wrong, which makes a Lower third built from raw output risky.
How RawClip handles it
Speech is transcribed to locate it and to protect it from being cut through, not to be shown on screen. Camera audio in action footage is too unreliable for captions to be generated from it without somebody checking the words.