Dirac, Audirvana and Audio Hijack Produce Weird High-Frequency Response, Why?

[Χ Ξ Σ]

I use both Audirvana and Audio Hijack to host Dirac Live Processor on macOS. While Audirvana offers a bit-perfect exclusive mode that shuts all sound but music, Audio Hijack enables system-level correction that also works for video content. I never thought these two hosts were any different until I saw the measurements.

To make things clear, let me explain the naming convention for my measurements. Take the first graph for example.

AudirThrough means when Audirvana is off or passing through the signal, and Dirac is also off. This shows the same effect as when Audirvana is turned off.
AudirDiracFull means Audirvana is on, Dirac is on and doing the full-range correction.
AudirDirac1kHz means Audirvana is on, Dirac is on and doing correction on the frequency response up to 1kHz. (From what I know Dirac is still doing phase correction above 1kHz and that cannot be adjusted.)
AudirDiracSwitch means Audirvana is on, and I am turning Dirac on and off, or I am switching between Dirac’s stored filters.
AudirDiracOff means when Audirvana is on, Dirac filter is off, but Dirac’s gain setting is still active.
Replacing Audir with Hijack and we got five more terms for what is happening when Audio Hijack is hosting Dirac.
8341+8341Filters means what it literally means. Later you will see the Genelec 8341 using filters designed for the Neumann KH80 and vice versa.

Off to the measurements. I use MMM and white noise.

First off we have 8341+8341Filters on Audirvana. As you can see in both the Log view and Linear view, I got some baffling results.

1. Audirvana on its own does not produce any anomalies in the high-frequency response.
2. When Dirac is on, the high-frequency response shows a zigzag. That is definitely not in the target response that I loaded. Where does that come from?
3. When Dirac is limited to 1kHz, the zigzag above 10kHz is still there. Shouldn’t the high-frequency response look like the AudirThrough response when no correction is turned applied?
4. A steep high-pass filter within the audible band? Really?
5. What is this notch? Pre-ring?

I thought Audirvana might be at fault, so I switched to Audio Hijack and found out that the hosting software actually affects the frequency response.

6. Still perfect.
7. Instead of zigzagging, the high-frequency response shows a slight roll-off, which, again, is not intended in my target response.
8. Again, limiting to 1kHz still affects all the way to 20kHz.
9. Same thing but more rolled off.
10. Same thing but more rolled off.

Then I thought perhaps the highborn Genelec was allergic to Dirac’s humble algorithm, so I switched up the speakers and the filters. I loaded an old set of Dirac filters designed for the KH80 and measured Audirvana’s response.

Well, on the 8341, when using Audirvana, the old KH80Filter behaved the same way as the newly designed 8341Filters.

Using Audio Hijack though, for some reason the notch disappeared. I rerun the remeasurements a few times and could not reproduce the notch here. But other problems are still there.

Finally, I measured the KH80 loaded with the set of Dirac filters designed for the 8341. The unintended roll-off and notch are back.

As for why these anomalies happened, I was hoping the technical experts on this site can give me some insight. By the way, before anyone suggests, I also have SoundSource. It behaved the same way as Audio Hijack did. Also in case anyone wants to see the smoothed response, here is one for the first graph.

 

[Χ Ξ Σ]

Maybe @Flak from Dirac can help me out here.

 

[txbdan]

Interesting. I use SoundSource to host DiracLive on my Mac. It seems to work well. I'd contact Rogue Amoeba as they're very responsive and helpful.

I don't remember the details, but I have an MQA enabled integrated amp that I had hooked up here to this computer once. I believe when I used SoundSource (its always running) but disabled Dirac, I could get the MQA decoder/validation light to come on meaning the file was passing through unmodified.. Of course turning on Dirac modifies the signal and MQA no longer lit up.

I also have Audirvana but I don't use it that often and don't use it to host the Dirac plugin. Maybe one of these days I can try to replicate what you're seeing.

 

[Χ Ξ Σ]

I have asked a technical expert/industrial insider about this, his guess was:

1. Though the frequency correction is limited, the phase correction is still active. Altering the phase will naturally alter the response in the frequency domain. No big deal.
2. Filters developed by Dirac might not be translated perfectly when going through another developer's specifications. From his experience, even importing a simple set of PEQ into different software can result in different measurments. Still nothing to swear about.

Later I did more Dirac calibrations and was able to produce a better looking high frequency graph on the right channel. Host is SoundScource. The left channel still has a tilt. But more often than not the tilt was presented.

 

[txbdan]

I'd break the test down a bit to isolate things. Maybe I missed it, but do you have:

1. Native output (no soundsource, no Dirac) vs Soundsource running but no Dirac loaded to test SoundSource's transparency. (All SS controls at 100% and no other effects of course)
2. SoundSource with no Dirac loaded vs Directed loaded but disabled in the Dirac plugin

 

[Χ Ξ Σ]

I didn't save measurements for SoundSource because during my test back then it behaved the same way that Audirvana did. Now I am using a different set of Dirac filters, the corrections are different, so they might look different now.