Capture at the highest resolution; you can always go lower
Quote
if word for word is just simply the best? You can always go lower resolution you can't go higher.
Episode 128: Fully Fingered (episode page, opens in a new tab) 00:11:27 (play audio from this timestamp, opens in a new tab)
Speaker-labeled
Summary
Arguing for word-level timestamps in JSON transcripts over line-based SRT: once you have the finest detail you can derive coarser versions, but never the other way around. Word-level timing also enables clipping by transcript on a phone.