Results of Automating Meeting Minutes with AI
I've been running a setup for a month now where I convert meeting recordings to text and have AI summarize them.
The clear upside is that the 30 minutes I used to spend organizing after a meeting has been cut to 5 minutes. But speaker diarization often gets it wrong, so in the end a human has to review it at least once.
In conclusion, semi-automation is the most realistic. Full automation is still too much to ask.
6 answers
Agreed, semi-automatic is the way to go. Full automation is still a fantasy.
What tool do you use for speaker diarization? I keep switching tools just because of that one thing.
I've been using something similar for about half a year now, and when it comes to speaker diarization accuracy, the mic setup is 80% of the battle. If you're picking up audio through a conference room speakerphone, even the best model will end up muddling everything together, so in the end the biggest win was using a separate mic per speaker, or for online meetings, setting it up so each person comes in through their own laptop mic. And if you lock the summary prompt to always split into three sections—'Decisions / Action Items / On Hold'—review time drops dramatically. FYI, I've got a script set up that deletes the recording files and outputs right after the meeting ends. Otherwise they just keep piling up in the folder and it becomes hell later lol.
Well, isn't saying 30 minutes became 5 minutes a bit of an exaggeration? Once you include review, it'll end up eating 15–20 minutes anyway. It's better than having someone organize it from scratch, sure, but the "5 minutes" figure seems to only count the automation part in isolation. I think we should re-measure based on the actual total time.
What did the security team say? If meeting recordings are sent out to an external API, ours got rejected outright. I wasted two months trying to find a model that runs only on our internal network.
That’s why I also just ask it to pull out only the decisions and toss all the rest of the chatter lol. That’s the real key.