The real-time question is the interesting one -- most enhancement models are tuned for offline processing where you can see the whole clip. Have you benchmarked latency on live audio, and does quality hold up on a genuinely bad source (laptop mic across a room) or mostly on sources that are already close to clean? Asking because we're building capture hardware and are curious how much of this is fixable after the fact vs. needs solving at the mic itself.
Very interesting, but, it seems it's offline only? I mean, one can't use it to speak through it on a meeting, correct?