We run a pretty strict "mute when not speaking" policy and it's working very well. I think giving users automated feedback about the quality of their sound is a much simpler problem to solve. Not as sexy as AI powered noise cancellation, but I think it would be much more effective in the short term.