Electric motor applications noise

I have been involved with the manufacturing and application of low (460 V), medium (4 kV) and high (13 kV) motors and am now on the driven equipment side of the coupling (large compressors), and the writing of IEEE and API equipment standards.
Many large end users endeavor to meet local or national noise exposure codes. And lots of the code setters are working to reduce noise exposure, with varying success.

In one experience (decades ago) in a California location, I was concerned about the 3000 Hp 3600 rpm motor driving a water pump, and meeting 85 dBA on my motor. As it turned out, the large motor was fine. The problem child turned out to be a 5 Hp motor driving a gear pump on the lube oil console (provided by the water pump vendor). It was noisier than my big unit.

Just think what a centrifugal compressor running at 10,000 rpm sounds like.

Typically, unmodified rotors spinning at higher speeds (ie 3600 rpm) have high noise at high frequencies (the most irritating part of dB A scale). 4 Pole, somewhat less, and so forth. Changing fans, baffels, etc. and etc. can help. Paddle wheel type fans are noisy, but bi-directional. Curved blade fans are quieter, but have to be uni-directional. Beyond that could be frame modification, part re-design (for weight, mass, harmonic performance.....)

In more situations, noise engineers are being employed to analyze noisy environments to engineer noise abatement beyond what the machine manufacturer can do. Building design changes, noise barriers, etc. (very expensive). For them, overall values (dBA) aren't good enough: they need not only octave band levels, but measurements taken at a wide range of locations spread throughout the installed environment (to the side(s), top, behind, etc.) for each piece of equipment. Models and computer constructs are nice, but the proof is in the actual installation and testing with microphones. It is the opposite of tuning a concert hall to get noise improvement: this is de-tuning to get noise reduction (and self-cancellation).

In the end, it comes down to money.
If a guarantee is involved with stiff liquidated damages, then lots of testing and certification on each serial numbered unit is likely to be necessary. (not to mention the cost of law suits....). Like the supply of a nuclear reactor: the reactor is free, it is the paper work that costs billions.
If extrapolation of values from typical machines using a quickie hand held meter is ok, then the impact is minimal.