Does it make sense to train kalman filter on unscaled (Timeseries-)data, to clean it?

I have Timeseries data to clean and I got the tip to use a kalman Filter.

My Question is, does it make sense to "fit" this kalman-filters parameters to an unscaled and unnormalized (therefore raw) dataset of timeseries, with the purpose to clean it?

Or will I get far more accurate results when normailzing, and then denormalizing it?