Auto EQ Analysis Doesn't Seem To Be Accurate

[vectorwarrior]

I often will try to 'capture' the frequency response of something by doing a 'before/after' analysis using one of Melda's auto-EQ's. One thing I've noticed is that capturing pink noise never creates a solid line, regardless of how I set up the spectrum analyzer, it always has wobbles. This makes me think the resultant auto-eq is going to be not totally accurate.

For example, I presumed that increasing the FFT size would get more accurate frequency analysis, but it's actually creating a more high-frequency output than lower settings. Also, setting things like 'super-resolution mode' and 'analytical smoothing', both of which are supposed to improve accuracy, don't seem to create a linear response either, sometimes creating less optimal results.

Try it for yourself: put a MNoiseGenerator into a track, drop a MAutoDynamicEQ (or whatever) afterwards and capture the noise. Note even after a minute of analysis the resultant capture is definitely not an average of pink noise and is quite wobbly, instead it should be pretty flat after a while (five minutes).

Thoughts?

[Faiky]

I think it's not a problem of analysis, it's due to the nature of the noise? And I think it's also because it's the band peaks that are measured and not their loudness.

[MeldaProduction]

It's most likely a combination of various inaccuracies. For instance the pink noise generator isn't perfectly flat (generating a pink noise actually isn't that easy). Then there's the lin->log conversion in the analyzer. Anyways nothing to worry about really.