→ all versions of this pipeline
Six attempts to locate this sound automatically have failed, and the last one made things worse. The problem is not the processing — it is that no span I have ever treated was confirmed by ear to contain the noise at all.
One of the spans I processed sits 1.9 seconds away from where you tapped. Another is 1.4 seconds off. If the span is wrong, then turning it down, replacing it, or filtering it are all equally pointless — and that is exactly what happened.
So: you draw the span. Play the window, drag the two handles until play just my selection contains the noise and nothing else, and confirm. Then I know where it is, and the fix becomes the easy part.
About the metallic whistling. That is musical noise — when a filter cuts some narrow frequency bands hard and leaves their neighbours alone, the survivors ring like little tones. It is a known failure of that whole family of processing, and the fix once the span is right is to turn the entire span down smoothly rather than band by band. No band-by-band cutting means no metallic artefact, by construction.