Mumble VS noise-suppression-for-voice

I think Sonobus is overkill. I suggest you look at a couple of relatively old-school gamer voice chat tools - Mumble or Teamspeak. Mumble is open-source and the connection is always encrypted, Teamspeak is commercial but the free tier should be fine for you - but you have to make sure to manually turn encryption on yourself. It has been a long time since I used either, so I don't know which is easier. Both of them require you to run their matching server software.

Mumble's latency is unbeatable imo, it's basically their main focus and shows.

The sticking point for me is the lack of persistent messages, something the devs strangely think is a privacy plus. Issue open since 2016: https://github.com/mumble-voip/mumble/issues/2560

If you drop out for a minute you won't have access to anything that was posted in chat, which makes it useless for anything other than voice only comms, that might suit some business purposes but I've always needed to post links or screenshots in chat during meetings.

There's any number of alternative chat applications available, like Element, Mumble, Teamspeak etc.

Mumble?

I think Mumble might fit what you're looking for. It's been a very long time since I've used it, but it seems to still exist: https://www.mumble.info/ - I've used previously for exactly what you're describing, events with lots of crew dispersed around and no budget for radios. I had it installed on an AP running OpenWRT so it was just a case of plugging that in and getting people to install the app and connect to it.

Along with the options already mentioned, if you're not into TeamSpeak, there is an open source alternative called Mumble which operates in the same manner. No internet required, and is supported on multiple platforms.

Yes, I think it pipes the default input (source) to the default output. I have a noise cancelling source and if I switch to it the loopback follows it and I hear the de-noised one

For linux: https://github.com/werman/noise-suppression-for-voice

Frankly, what I hear is very similar to the results of classic spectral denoising, even with the characteristic artifacts (for Linux, there's Noise Repellent [1] available for advanced spectral denoising; there's also a ton of commercial spectral processors).

The demonstration could use more random background noises to separate it from spectral processors, and more varied vocabulary to separate it from RNNVoice [2] which tends to suppress breath and parts of sibilants, making the sound unnatural. The latency is also important - is it as low as in RNNVoice? What about the CPU load?

[1] https://github.com/lucianodato/noise-repellent

[2] https://github.com/werman/noise-suppression-for-voice

Nice, hopefully you or somebody else will turn this into an easy to use PipeWire plugin, so I can replace noise-suppression-for-voice.

I use this plugin to remove the sound of my air conditioner, which can be quite loud: https://github.com/werman/noise-suppression-for-voice

This comes very close to RTX Voice while being multiple times more efficient. Not as good as RTX Voice but I prefer much lower HW usage.

bonus points if you use equalizerAPO + a VST plugin for noise suppression if you really need to suppress noise.

I like the RNNoise plugin so much I ended up using it in my audio tracks too when doing vocals.

equalizerAPO + https://github.com/werman/noise-suppression-for-voice