julian since you asked!

I'll need a bit more than upvotes here

thisismissem hah yeah, I was doing dinnertime and didn't have the bandwidth to answer, one sec

thisismissem hedgedoc or hackMD worked surprisingly well. Before I had a Google Docs sheet and participation was low because of the access control barriers.

I would not be opposed to an AI scribe, and I actually have access to one with my Google Workspace sub, but I think that comes with a lot of baggage that precludes its use.

The worst part of scribing your own notes is that you can't participate. All my notes for sections I lead don't ever get captured fully, and I have to do it post-meeting... about half the time it doesn't happen at all

As mentioned offline, perfectly fine to use dmitri@social.coop's HedgeDoc instance.

Yeah, in the past I've gained consent to record a call & use a local model to transcribe which I then edited, but it was time consuming.

In the T&S taskforce grant I've applied for money to pay a scribe for their time

I think I'll give it a try. I did some light research and am going to give Whisper.cpp and some Ollama models a try. I'm not sure whether they'll have the ability to summarize and distinguish voices, but let's see... I'll report back in this thread.

I have been informed that there is an (outdated?) Etherpad instance at https://pad.w3.org that you authenticate with via Basic Auth using your W3C Account credentials.

Okay, I spent the better part of an hour playing around with it (after compiling whisper.cpp and going down a weird rabbit hole — thanks Gemini!)

I recorded some test audio, had it transcribed, and passed it over to ollama running llama3.1, with a prompt to summarize the meeting into bullet points and action items. It seemed to go over well, although I don't quite know how it will handle digressions, asides, and hour-long meetings

It might be we can adjust the prompt so that it isn't summarizing so much as efficiently condensing meeting discussions.

Definitely worth a shot if only to reduce the burden of scribing while chairing a meeting!

julian I used vibe.app when I did transcription, and it kinda butcher audience a few months ago and "fediverse" became like "fetty verse", but otherwise was okay.

Whisper.cpp is surprisingly good. It is getting 99% of it... missing things it doesn't know about (like PyFET instead of Piefed), but it definitely knows what activitypub is, so that's handy.

Thanks for pushing me (even if indirectly) to do this. Looking forward to trying it out for real.