A 50Hz notch filter is a better tool for that, as it will do less collateral-damage to the speech than NR.
The nature of human-hearing is that the perceived frequency-content changes depending on volume …
Your “after” is considerably louder than the “before”, (~10dB louder).
So you’ll have to adjust the equalization to make loud version sound the same as the quiet version
(You may also have to use a de-esser).
There are free equalizer plugins which work in Audacity in Windows,
which behave like a real hardware equalizer that you can adjust in real-time when the audio is playing …
TDR Nova | Tokyo Dawn Records *
GVST - GMulti *
( * can equalize & de-ess at the same time).