@dave, I've created a PR that implements voice message posting. Maximum duration is 4 minutes which should fit under 2MB limit.
The PR adds browser voice recording with MediaRecorder, optional live transcription, Mediabunny MP3 conversion, cancellable media cleanup, enclosure-backed audio players, and nightly orphan collection.
Also include a repeatable localhost development setup with SQLite, CORS-aware client serving, local config bootstrap, and documented npm 12 native-build recovery.
PS: I am pretty sure this PR is not a hallucination.
PPS: I think you asked me for something like this 6yrs ago. Better late than never.
PPPS: Quality of transcription is horseshit which is why I typical used a embedded model but I figured you wouldn't like rss.chat loading models weighing hundreds of MBs.
Don, thanks for this idea, we can't use it here at this time, for this reason:
I want to keep this site as the basis for what we're doing overall, and while voice in a conference system like this makes a ton of sense, it isn't a core feature.
But there seem to be a few forks of RSS.chat out there, and they're adding features at a pretty rapid rate. Why don't you try this idea out with one of them? Then we could see what people do with it, without changing what the reference version does.
That's fine. I consider voice posting to be a core feature that grants rss.chat a fluid multiuser async voice chat experience. You should try it out if you haven't because it feels different than how one might imagine. A strong whiff of magic sauce.
I've closed the PR but, if anyone wants to try it, it's in the "voice-post" branch of my fork.
I've merged my brain into main branch of my fork and renamed the fork as rss.voice to give it a distinct, ur, voice. So the new link is: https://github.com/donpark/rss.voice
