Friendica
Sean Randall
Sean Randall

Sean Randall

cachondo@defcon.social

Sean Randall

cachondo@defcon.social
I work for a national telecommunications company to help ensure accessibility across web and mobile. I code for leisure, fun and accessibility.
I geek out on sci-fi movies and am constantly reading.
I enjoy the odd game here and there.
I watch comedy with my wife, horror with my daughter, and more serious scary stuff with the dog.
I listen to classical when working, and scare off cats to pop, showtunes and country when tidying.
ActivityPub
2025-06-07 15:59:55 2025-06-06 09:55:26 2025-06-06 09:55:25 7926815

Sean Randall
Sean Randall
mastodon - Link to source

Sean Randall

3 months ago • •

Sean Randall

3 months ago • •


For day 6 of #AudioMo, here's an AI attempt at audio-describing a long point of tenis.
Unfortunately I've had to butcher the description for time, but the words are all its own.
https://files.defcon.social/dcsocial-s3/media_attachments/files/114/635/814/597/760/030/original/1b66490524a933f8.mp3
#audiomo
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Sean Randall
mastodon - Link to source

Sean Randall

in reply to Sean Randall • 3 months ago • •
and, for those of you who can see, here's the actual video. #AudioMo
https://files.defcon.social/dcsocial-s3/media_attachments/files/114/635/818/462/994/974/original/1f14deeb3c378971.mp4
#audiomo
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Dave Taylor
mastodon - Link to source

Dave Taylor

in reply to Sean Randall • 3 months ago • •
So, whatever you are saying is so quiet that it's almost inaudible, and you just put us in a player that doesn't show any tags. In saying that, I bet we could get some really good info about tactics from AI that sighted people wouldn't even think to mention, even radio commentators
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Jakob Rosin
mastodon - Link to source

Jakob Rosin

in reply to Sean Randall • 3 months ago • •
Honestly, If I wouldn't know its an AI... wow.
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Jakob Rosin

Sean Randall
mastodon - Link to source

Sean Randall

in reply to Jakob Rosin • 3 months ago • •
@jakobrosin I imagine that it won't be long before we can ask an LLM to fit its description around unspoken parts of a video. Especially if it has the script, as well, there's real scope for it to do a good job at this sort of thing.
@Jakob Rosin
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Jakob Rosin
mastodon - Link to source

Jakob Rosin

in reply to Sean Randall • 3 months ago • •
Very likely. Curious, what LLM did you use for generating the description? Did you feed it the video or just sequence of stills?
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Jakob Rosin

Sean Randall
mastodon - Link to source

Sean Randall

in reply to Jakob Rosin • 3 months ago • •
@jakobrosin Gemini. I gave it the entire video and asked for a description in very broad terms.
tempted to give it a scene from a movie or something and ask it for a subtitle file around the nondialogue bits, see how it handles it.
@Jakob Rosin
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Jakob Rosin
mastodon - Link to source

Jakob Rosin

in reply to Sean Randall • 3 months ago • •
It looked like it used both the image and audio as its imput. Thanks for the tip, I didn't know gemini can take video as an imput now. Maybe I need to broaden my llm subscriptions
  •  Languages
  •  Search Text
  •  Share via ...
in reply to Sean Randall

Kevin R Jones
mastodon - Link to source

Kevin R Jones

in reply to Sean Randall • 2 months ago • •
What if you tried to speed up the voice, would that help keep it in sync? That was a cool project you did, I wouldn’t have thought of that.
  •  Languages
  •  Search Text
  •  Share via ...
⇧