The easy way to build a meeting note taker is to build a bot. You get a clean audio feed, per speaker, handed to you by the platform, and the whole problem of desktop audio capture disappears.
It also means a stranger appears in the participant list of every call your customer takes. We tried it. People hated it, and they were right to. The bot changes the conversation, and nobody in the room agreed to it being there.
So Expo reads system audio instead. This is harder. It means a native adapter per platform, permission flows that differ by operating system, and no free speaker labels. What it buys is that nothing shows up in anyone else’s client.
The trade we made is that the person running Expo is responsible for telling the room. We make that as easy as possible by keeping the capture state on screen the entire time, but we do not pretend the software solves consent on its own.