Post #3604287

@grimalkina@mastodon.social My cents as a deaf person who often benefits from transcriptions: pay as much attention to the latency as to accuracy. Modern STT software is pretty good on accuracy (i.e. word error rate) but even industrial ones like Zoom and Teams can be so bad (500+ ms on closed captioning) it’s worse than no transcript at all. General guidance is captioning should be <200 ms to enable real time engagement for people who absolutely need them.