As using “voice as an input” is one of the grails in VR, in my opinion… I would suggest looking at Voice Attack. Something similar in functionality for using it as a voice input. For voice communications, well, steam voice chat would have that. However, it would need to go sending client -> server -> receiving client(s).
The sending client would need to either:
A) When PTT button is pressed, then “Microphone Listen” with Text input “say something here” then do XYZ
B) If Voice Activated option is enabled, then Microphone Listen should always be turned on
Performance shouldn’t be that bad, as games that would implement this, would have their player’s download / use teamspeak / voice attack / skype / whatever, so the overhead cost should be roughly the same.
Some of the research documents I found when I was completely failing to make a plugin for voice recognition:
https://azure.microsoft.com/en-us/services/cognitive-services/
TBH, I cannot see Epic doing this either. It’s actually not for, well, sort of not for, simulations, but more for making an extremely accurate calculator (for the most part)… quantum physics / astronomy math and all. Basically, it would be for 1 person in the world; me. And I have no idea how to program… soo 
edit: I actually found a short cut to this precision level, and its much more efficient than using that big of float precision’s. So much more efficient, that from testing, I can run it on my LG Zone3 phone. Need peer review on it, of course, but eh… I’m not at that point yet in that project, which is on the shelf atm.