Mastodon Feed
Reblogged by cstanhope@social.coop ("Your friendly 'net denizen"):
edsu@social.coop ("Ed Summers") wrote:
Along with everyone else (bah AI) we've been experimenting with Whisper for transcribing media that lack transcripts. The thinking being that any transcript is probably better for accessibility? (we'll see)
We noticed that when you tell Whisper to generate word timestamps the JSON file it generates also includes confidence levels for each word [0-1].
I thought it might be useful/fun to be able to view the JSON transcript with confidence levels colored, so here is https://edsu.github.io/whisper-transcript/