We host a public mumble server - it's at
mumble.cyberia.club on the default port.
1. Download and install mumble (github.com/mumble-voip/mumble/releases)
2. Connect to mumble.cyberia.club
In order to access our private channels, you may use the following
Server -> Access Tokens -> Add
There's a few options being considered, including hiring a professional transcriber. We need something ready by January(?) 2023:
that's the idea here i think. we just need something that works that we can use for our mumble meetings, and using a cloud API is a tradeoff we'll have to make for our own sanity in the short term. once we have that we can ponder ideas of more "privacy friendly" solutions that don't ship audio data off to the big data eldritch monster, but for now a working proof of concept is our goal.
it's missing the filters and ring buffer found in the stream example, which should greatly enhance transcription quality if implemented, but a lot of CPU is required, and GPU acceleration has yet to be explored.
hey so, after being prompted into remembering this project I bumped into the code again and now it's quite a bit nicer! in that it actually works well that is. it's using deepgram.com which I discovered via a (very nice) youtube shill and it was pretty convenient to use.
on sign-up you get, $150 of credits, and currently it's somewhat wasteful with them, but that's okay. I think it might incidentally be not too bad, but it sends every audio packet it gets without any VAD.
so the congress meeting synthesized itself today and it was quite nice! the transcription bot kept on crashing but was somewhat usable to reese! 💖 (as far as ckie could tell)
- [x] the Interim Results deepgram feature was turned on and afterwards they told ckie that they'd rather it be off for next time.
- [ ] also the Mumble chat was made useless for other purposes by how spammy it was.
- [x] the bot kept on crashing because of mysterious programming reasons