◎
It listens
The audio becomes text with a timestamp on every word, so a match points at a second rather than a paragraph.
Built by Krzysztof. Describe what you're after in a recording and IRIS gives you the timestamp. Not a transcript to read through, the minute to jump to.
The idea
Stop dragging the scrubber. Ask for the moment instead.
You remember someone explained the thing. You just don't remember when. IRIS listens to the whole recording, works out what your description means, and hands you the timestamps. Roughly describing it is enough, you don't need the exact words.
How it works
All of it runs on open models, and it can run on your own machine, which matters when the recording shouldn't leave it.
The audio becomes text with a timestamp on every word, so a match points at a second rather than a paragraph.
You search by idea, not by keyword. "The bit about why the deadline moved" finds it even if nobody said the word deadline.
When words aren't enough it checks the picture. Useful when the thing you want was on a slide and nobody read it out loud.
In practice
Upload, type roughly what you're after, get back the points to jump to. That's the whole loop. No scrubbing, no guessing which quarter of the recording it was in.
Because it runs on open models and can stay local, a recording you'd rather not upload anywhere doesn't have to go anywhere.
Good for
IRIS
Bring something long and a rough idea of what you want out of it. IRIS points at it.