Methods and workflow
Real-time transcription
Transcription that happens while the speaking happens, with one to three seconds of delay.
Real-time transcription is transcription that happens while the speech is going on. The text arrives with one to three seconds of delay.
The counterpart is batch transcription, where the whole recording is processed afterwards.
Why the delay exists
The system has to hear a little way ahead to interpret the words in context. Without that small buffer, every word would be chosen without context, and the result would be noticeably worse.
The trade-off is hard: a shorter delay gives faster text and more errors. A longer delay gives better text, but becomes tiresome to follow for someone reading along.
That is why you often see the text adjust itself a couple of times before it settles.
What it costs in accuracy
A system that processes the whole recording afterwards can see what was said two minutes later and use it to interpret an unclear word. A real-time system only has the past. The difference shows most on proper nouns and jargon.
What it is used for
Captioning for accessibility, lectures, webinars, live broadcasts and dictation. Everything where somebody is meant to follow along as it happens.
If what you want is a summary at the end, batch is the right choice.