My results:
Starting with white noise generated with amplitude 0.5
Before B&T:
Peak: -6.021 dB
RMS: -10.80 dB
Analyze > Plot Spectrum (Size: 2048): Flat, approx -36 dB
B&T settings:
Bass: -12 dB
Treble: 0 dB
Volume: 0 dB
Link volume: Off
After B&T:
Peak: -3.435 dB
RMS: -11.22 dB
Analyze > Plot Spectrum (Size: 2048): Curved, approx -48 dB on left, -36 dB on right.
There’s a reason why the peak level increases. It’s not a bug.
White noise is random numbers where each sample value “x” is a random number between a max and min value.
For white noise amplitude 0.5, each sample value is a random number: -0.5 > x < 0.5,
with equal probability for any value within that range.
In other words, the 0.5 peak white noise is random number constrained to the range -0.5 to + 0.5.
The -12 dB bass (shelf) filter has little effect on the overall (RMS) level, because white noise has relatively little energy in the bass frequencies (hence it sounds like “sssss” rather than “urrrrr”).
However, the filter does jiggle around all of the sample values, and they are NOT constrained to the range -0.5 to +0.5, so some of the sample values end up outside of that range. Thus, given that “peak” level is measuring the outliers (max absolute sample values), the peak measurement goes up.
Does that make sense?
If you try the same test with Pink noise, you will see more “expected” results, because Pink noise has more energy in the bass frequencies and is less tightly constrained.
I note that in this post: Mastering - #28 by TJB
they are boosting both bass and treble, with a small reduction in the overall volume. That will push up the low and high frequencies, and pull down the middle. The overall effect on peak level cannot be predicted, but it is likely to be slightly increased with the settings they are using.