I'd argue it's because there's a lot of problematic content that gets shared in text that just isn't really much of an issue (or isn't viable to detect) in audio/video.
I'd argue the opposite somewhat: there's a lot of problematic content that's an issue with audio/video, but like you said, it's not viable to detect at scale, so it's better to close the door.
The cynic in me agrees with you here - this is likely a way for them to go "oh no, we couldn't see that information, it's *encrypted* so we have no liability, legal or otherwise, to stop any sort of abuse on our platform since we can't see it"
Well, this is why Signal is fine while the Telegram boss is in jail. As long as you haven't done anything illegal (and aren't explicitly trying to enable illegal activity), it's perfectly fine to just say "we can't do this." I'm really for this; being able to inspect users' data should be a liability.
I don't feel convinced of this takeaway, at least in the context of being applied across the board.
I help administer a semi-large, public studygroup community that sees its share of trolls and the like joining the channels and causing disruptions (up to and including exposing themselves and masturbating/helicoptering) for shock value, etc.
If anything, I find Discord's moderation tools for server administrators painfully lacking. Discord is not Signal.
I would have liked to see this in some form closer to an assignable privilege to send out/upload E2EE data granularly grantable to server regulars, while new people start out without the privilege.
This press release going into cool technical details in order to tout E2EE and namedropping one of the most reputable consultants in the biz feels a little tonedeaf.
more sensibly, and of which i would be really receptive to (as a server/guild administrator), granular setting on a per-channel basis.
of course, this sentiment largely takes for granted that there is any open-facing mission on Discord's part to facilitate community moderation; i definitely tend to lean privacy-first in general.