There is a significant level difference between octave smoothed and the raw data

[3ll3d00d]

Applying fractional octave smoothing lifts the level significantly, some of this is to be expected but the difference is really quite large so it needs further investigation to determine if this is correct and/or can be accounted for

[aron7awol]

This seems to vary with the fraction of the octave, which may suggest a math error somewhere? (I admit I haven't looked at the method of smoothing so it's a shot in the dark, but figured I'd share what I noticed in my very limited testing)

[3ll3d00d]

that is to be expected because the size of each bucket is changing, basically smaller fraction means smaller bin means less signal (energy) per bin