Good point, piezos do also generate voltages when deformed, so this could conceivably be run in reverse... Microphone arrays are already used for directional sound detection, so this could have fun implications for that application given the claimed density of microphones.
You could likely do both at the same time.
The "microphone" would be the measurement of interference between speaker demand and actual response.
Would a piezo crystal, mounted behind a screen and the glass panel atop it, have enough sensitivity to capture recognizable speech?
If you put a dualsense controller atop of glass and send sound through it to the back-left or back-right channels (controls the haptics) you definitely hear it, so I presume yes.