I recently encountered an article that described how for the last few decades, the UK has been recording the very small changes in frequency of its electricity generation system.
Then, when there are situations where someone records a video and claims that it was recorded at a particular time, experts extract the faint electricity hum, and compare it to the records for the claimed time, in order to see if it matches, to validate the claim. If it is an unedited video then the experts can find out when it was really recorded. If it was edited together then experts can detect the edit points and the actual recording time of each segment.
Now, for those experts, whatever is in the video, apart from the electronic hum, is noise to be filtered out. Whereas to everyone else, if they notice the electronic hum at all they would want it suppressed.
This establishes that there cannot be any objective distinction between what is "noise" and what is not. "noise" is a form of information.
Therefore your task cannot be accomplished.
Now, what might be possible is for you to create objective rules describing what should be removed or retained. Not "noise" but instead something like "bandpass between 12 Hz and 8000 Hz and remove between 59.5 and 60.5 Hz"