Log-based normalization for frequency params

[tompkinsguitar]

The normalize function is

(val - min_val) / (max_val - min_val)

However, it might make sense to also do log10 (or some related scale) to the parameters involving frequency ranges (and maybe q factors?) before normalization.

For example, humans perceive the change from 100Hz to 400Hz as the same relative amount as from 400 to 4000. So doing a log10 scaling then normalization for frequency would have the model output scale match human perception and might be easier to train.