[castleofargh]

this is beyond veritas and REW, just waste a night googling about loudness measurements standards and what exists to estimate perceived loudness(which has little to do with actual loudness). in the end it will be your preference and not much else.
you've set REW for 1khz tone, that's the only place so far where you know for a fact that the values are correct. the microphone doesn't have a flat response(most professional microphones would come with an individual calibration, making it all much easier). the IEM also won't be flat, the er4 for example isn't the loudest at 1khz.  so we don't really have what it takes to start worrying about a few db or take what we measure at face value. this is low fi measurement, we have to go with what we can do for the money.
 
 
 
in case it's not clear, Zpeak stays at the highest peak value it gets, just like LAF max doesn't go down over time. it's a little soon to go all paranoid on values ^_^.
and Z is no weighting so if you get some good bass(where often the max amplitudes will be in music), plus your IEM doesn't roll off like crazy in the low end, then it will show.

 

[csglinux]

I've got a little question for the experts and headfi veterans...
 
When taking a measurement in REW, by default, it does no smoothing, but it has the usual 1/nth-octave smoothing options, if you want to add them later. Does that mean by default REW is creating power-spectral density? I presume it can't just be narrow-band output, because that would change with sample length (Parseval's theorem telling us the the power in the time-domain has to equal the power in the frequency domain). The REW manual seems a bit spartan on this topic. Any thoughts?

 

[bartzky]

@bartzky
, have you ever used .req files with REW for EQ/correction? These are binary files, so I've having fun editing them Is there an easy way to make up and import parametric EQ files? 

I've been fiddling around with REW's EQ a bit. An easy way to export EQ settings is the txt file format. It can easily be imported to programs like Equalizer APO.

 

[csglinux]

Thanks @bartzky - I actually figured that out too. I've been averaging some measurements in order to generate an ASCII compensation curve for AudioTool. I'm messing about with @crinacle's trick of playing white noise into a smartphone app. It's not bad, but with the exact same Vibro Veritas mic, AudioTool is a little bit off in the highs and lows. But a lot of that can be "fixed" with compensation curves  BTW, I didn't have any luck at all with the iPhone apps. I tried FFT Plot, but it was a disaster for me.
 
I've spent some time over the last couple of weeks emailing back and forth with John (REW developer). I need to send him a larger donation at some point as his tech support is beyond incredibly awesome. I've written FFTs, FIR and IIR filters and stuff in the past and thought I knew it all. After several dozen emails from John, it's now become crystal clear as to just how little I actually know about DSP BTW, REW doesn't output PSD - it's SPL (in Hz) and octave smoothing is only that - it's just a finite or infinite impulse response smoothing, not the traditional octave-band output I'm used to. It relies on you calibrating your entire setup to get the correct SPL. Also, octave smoothing will round off extrema, so one has to be a little bit careful when comparing A vs B. I've used REW to create 1/3rd octave compensation curves for AudioTool that take me straight to diffuse-field compensated output. It's slower and not as accurate as the log f sweep that REW does, but it's impressively close for a portable setup, and AudioTool also lets you export the frequency measurement as an ASCII file. So you can basically run it twice, once with the compensation curve and once without (or with only a mic/smartphone-only compensation curve), and generate both raw and diffuse-field compensated curves which you can then send to any other software. Kudos to @crinacle for figuring out a really nice portable measurement setup!
 
If anybody's interested, this paper gives a great overview of the various different ways of computing the transfer function and explains the errors inherent in using a white noise signal:
 
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.118.7473
 
I can't post the actual link to the paper itself, because it's copyrighted, but if you google hard enough, you'll find the whole thing

 

[csglinux]

Just as an example, here's a comparison of the Vibro Veritas mic being used with AudioTool and crinacle's white-noise signal (red) vs the exact same IEM (T-Peos Altone 200) measured with a 512k log sweep in REW using the exact same coupler. No calibration curves of any kind with either setup (you can click for a larger image):
 

 
The white noise seems to lack a bit in the lows and highs (I found that even with a loopback without any mic), but the curve shapes are otherwise pretty close.
 
P.S. See my AC power-supply glitch in REW?!