In my completely non-expert opinion, I think it would also be a hard dataset to normalize across mic sensitivities/auto gain control/time of day/weather/phone orientation/etc., even if you got a billion people to submit readings.

Still, there have been previous attempts at this:

2013: https://www.technologyreview.com/2013/10/22/112756/noise-pol...

2019, with calibration: https://www.sciencedirect.com/science/article/abs/pii/S00036...

2023, iOS only in the USA: https://deohs.washington.edu/noise-across-america-study

There's a bunch more on Google Scholar, actually.