You would have to use Microsoft Speech Platform 11 to support multiple users.
The built-in Microsoft Speech Recognition engine is designed to be trained by a single user(with a single microphone setup), so it likely won't do well in any other scenario.
To get the audio to work, you'd have to use something like
a virtual audio cable to present discord output as a recording device, and a
virtual mixer to mix your microphone input with TTS/sound file output.
However, VoiceAttack plays sound via the default audio playback device, so I'm not sure you can separate the ingame sound so it doesn't play over Discord.
If the AI didn't have to interact with your game, the "simplest" way to set this up would be to run it standalone in a virtual machine, with its own connection to Discord(You could communicate between the virtual VoiceAttack instance and the one on your physical machine by plugin and pass commands from one to the other, but that's not a simple thing to set up, by any means).
*I'm not affiliated with, or even have used the software linked, but it's what I know exists for this purpose.